Glossary of Artificial Intelligence

A* Search

A graph exploration and route-finding algorithm widely applied in various areas of computer science because of its thoroughness, optimality, and efficient performance.

‍

Abductive Logic Programming (ALP)

A broad knowledge-representation paradigm that facilitates problem-solving in a declarative manner using abductive inference. It eases conventional logic programming by permitting certain predicates to remain partially specified, designating them as abducible predicates.

‍

Abductive Reasoning

A type of logical reasoning that begins with an observation or a series of observations and then attempts to determine the most straightforward and probable explanation. Unlike deductive logic, this method produces a likely conclusion without definitively confirming it. Also known as abductive inference or retroduction.

‍

Ablation

The elimination of a specific element within an AI system. An ablation study seeks to assess the impact of that element by extracting it and subsequently evaluating the system's performance to determine its contribution.

‍

Abstract Data Type

A formal mathematical framework for data types, wherein a data type is characterized by its functional behavior from the perspective of a user. This is defined in terms of permissible values, applicable operations on data of this type, and the expected behavior of those operations.

‍

Abstraction

The procedure of eliminating physical, spatial, or temporal specifics or attributes in the examination of objects or systems to focus more precisely on other aspects of significance.

‍

Accelerating Change

An observed acceleration in the pace of technological advancement, potentially indicating more rapid and significant transformations in the future, which may or may not be accompanied by equally substantial social and cultural shifts. 

‍

Accuracy

Accuracy is a measurement metric in binary classification and is determined using the formula: (True Positives + True Negatives) / (True Positives + True Negatives + False Positives + False Negatives).

‍

Actionable Intelligence

Data that can be utilized to aid in decision-making.

‍

Action Language

A formal notation for defining state transition systems, frequently utilized to construct precise models of how actions influence the surrounding environment. Action languages are widely applied in artificial intelligence and robotics, where they depict the impact of actions on system states with the passing of time and can be employed for automated decision-making and planning.

‍

Action Model Learning

A branch of machine learning focused on developing and refining a software agent's understanding of the outcomes and prerequisites of actions it can perform within its surroundings. This knowledge is typically encoded in a logic-based action description language and serves as input for automated planning systems.

‍

Action Selection

A method of defining the fundamental challenge faced by intelligent systems: determining the next course of action. In artificial intelligence and computational cognitive science, the action selection problem is commonly linked to intelligent agents and animats, which are artificial entities that demonstrate intricate behaviors within an agentic environment.

‍

Activation Function

In artificial neural networks, a node's activation function determines its output based on a given input or collection of inputs.

‍

Adaptive Algorithm

An algorithm that adapts its behavior during execution based on a predefined reward system or evaluation criterion.

‍

Adaptive Learning

Adaptive learning is an approach that leverages data-informed instruction to modify and customize educational experiences to suit the unique requirements of each learner. Adaptive learning platforms can monitor information such as a student's advancement, involvement, and achievement, utilizing this data to deliver individualized learning experiences.

‍

Adaptive Neuro Fuzzy Inference System (ANFIS)

A type of artificial neural network derived from the Takagi–Sugeno fuzzy inference system, developed in the early 1990s. Since it merges neural networks with fuzzy logic concepts, it has the potential to harness the advantages of both within a unified framework. Its inference mechanism consists of a collection of fuzzy IF-THEN rules with the ability to learn and approximate nonlinear functions. Consequently, ANFIS is regarded as a universal estimator. For improved efficiency and optimization, the best parameters can be determined using a genetic algorithm. The neural network is also referred to as an adaptive network-based fuzzy inference system. 

‍

Admissible Heuristic

A heuristic function is considered admissible if it never predicts a cost exceeding the actual minimum cost required to reach the goal. In other words, the estimated cost to reach the objective is always less than or equal to the least possible cost from the current position along the path.

‍

Agent Architecture

A framework for software agents and intelligent control systems, illustrating the organization of components. The structures utilized by intelligent agents are known as cognitive architectures.

‍

AI Accelerator

A category of microprocessors or computing systems engineered as specialized hardware to improve the performance of AI tasks, particularly artificial neural networks, computer vision, and machine learning. 

‍

AI-Complete

In artificial intelligence, the most challenging problems are informally termed AI-complete or AI-hard, indicating that their complexity is on par with solving the fundamental AI challenge, creating machines with human intelligence, also known as strong AI. Labeling a problem as AI-complete suggests that it cannot be resolved using a straightforward, specialized algorithm.

‍

AI Ethics

AI ethics encompasses the concerns that AI stakeholders, including engineers and policymakers, must address to guarantee the responsible development and deployment of the technology. This concept involves designing and applying systems that promote a safe, reliable, impartial, and environmentally sustainable approach to AI.

‍

Algorithm

An algorithm is a set of instructions provided to an AI system to carry out a task or resolve a problem. Common types of computer algorithms include classification, regression, and clustering. 

‍

Algorithmic Efficiency

A characteristic of an algorithm that pertains to the quantity of computational resources it consumes. An algorithm must undergo analysis to assess its resource utilization, and its effectiveness can be evaluated based on the consumption of various resources. The efficiency of an algorithm can be likened to engineering productivity in a recurring or ongoing process.

‍

Algorithmic Probability

In algorithmic information theory, algorithmic probability, also referred to as Solomonoff probability, is a mathematical approach for allocating a prior likelihood to a specific observation. This concept was devised by Ray Solomonoff during the 1960s.

‍

AlphaGo

A software application designed to play the board game Go. It was created by Google DeepMind, a subsidiary of Alphabet Inc., based in London. AlphaGo has multiple iterations, including AlphaGo Zero, AlphaGo Master, and AlphaGo Lee, among others. In October 2015, AlphaGo became the first AI-powered Go program to defeat a human professional Go player without handicaps on a standard 19×19 board.

‍

Ambient Intelligence (AmI)

Digital surroundings that are perceptive and reactive to human presence.

‍

Analysis of Algorithms

The assessment of an algorithm's computational difficulty, specifically the quantity of time, memory, and other resources required for execution. Typically, this entails identifying a function that correlates the size of an algorithm's input to the number of operations it performs, its temporal complexity, or the amount of storage it consumes and its spatial complexity.

‍

Analytics

The identification, analysis, and conveyance of significant trends within data.

‍

Anaphora

In linguistics, an anaphora is a pronoun used to refer back to a previously mentioned noun. For instance, in the sentence, "While Paul didn't like the appetizers, he enjoyed the entrée," the word "he" serves as the anaphora.

‍

Annotation

The method of labeling linguistic data by recognizing and marking grammatical, semantic, or phonetic components within the language data.

‍

Answer Set Programming (ASP)

A type of declarative programming designed for tackling complex, mainly NP-hard, search challenges. It is founded on the stable model, or answer set, semantics of logic programming. In Answer Set Programming (ASP), search problems are transformed into the task of computing stable models, with answer set solvers, tools that generate stable models, being utilized to execute the search process.

‍

Ant Colony Optimization (ACO)

A stochastic method for resolving computational challenges that can be transformed into the task of identifying optimal routes within graphs.

‍

Anytime Algorithm

An algorithm capable of producing a legitimate solution to a problem, even if halted before reaching completion.

‍

Application Programming Interface (API)

An API, or application programming interface, is a collection of protocols that define how two software applications communicate with one another. APIs are typically written in programming languages like C++ or JavaScript.

‍

Approximate String Matching

The method of identifying strings that closely resemble a specified pattern rather than matching it precisely. The challenge of approximate string matching is generally categorized into two sub-problems: locating near-matching substrings within a given string and retrieving dictionary entries that approximately correspond to the pattern.

‍

Approximation Error

The deviation between a precise value and an estimated approximation of it.

‍

Argumentation Framework

A method for handling disputed information and deriving conclusions from it. Within an abstract argumentation framework, foundational data consists of a set of abstract arguments that, for example, symbolize facts or assertions. Disputes between arguments are depicted through a binary relationship within the argument set. In practical terms, an argumentation framework is illustrated as a directed graph, where nodes signify arguments, and arrows denote the attack relation. Several extensions of Dung's framework exist, such as logic-based argumentation frameworks and value-based argumentation frameworks.

‍

Artificial General Intelligence (AGI)

A form of artificial intelligence that equals or exceeds human intellectual abilities within a broad spectrum of mental tasks.

‍

Artificial Immune System (AIS)

A category of computationally intelligent, rule-based machine learning frameworks influenced by the mechanisms and functions of the vertebrate immune system. These algorithms are generally designed to replicate the immune system's adaptive learning and memory capabilities for problem-solving applications.

‍

Artificial Intelligence (AI)

AI, short for artificial intelligence, refers to the emulation of human cognitive processes by machines or computer systems. AI can replicate human abilities such as communication, learning, and decision-making.

‍

Artificial Intelligence Markup Language

An XML-based language variant designed for developing intelligent software entities that process natural language.

‍

Artificial Narrow Intelligence (ANI)

AI can address specific issues, which is referred to as artificial narrow intelligence. For instance, a smartphone can utilize facial recognition to detect images of a person in the Photos app, but that same system is unable to recognize sounds.

‍

Artificial Neural Network (ANN)

Often called a neural network, this system comprises a network of interconnected nodes or units that roughly emulate the processing capabilities of the human brain.

‍

Association for the Advancement of Artificial Intelligence (AAAI)

A global, nonprofit, scientific organization dedicated to advancing research in and the ethical application of AI. AAAI also seeks to expand public awareness of AI, refine the education and training of AI professionals, and offer recommendations to research strategists and funding bodies regarding the significance and prospects of contemporary AI advancements and emerging trends.

‍

Asymptotic Computational Complexity

In computational complexity theory, asymptotic computational complexity refers to the application of asymptotic analysis to estimate the computational demands of algorithms and computational challenges, typically linked to the use of Big O notation.

‍

Attention Mechanism

A machine learning-based attention mechanism is a system that emulates human cognitive focus. It assigns soft weights to each word, specifically to its embedding, within a contextual window. This process can be performed either concurrently, as in transformers, or sequentially, as in recursive neural networks. Unlike hard weights, which are pre-trained, fine-tuned, and remain fixed, soft weights can dynamically adjust during each execution. Transformer-based large language models utilize multiple attention heads.

‍

Attributional Calculus

A reasoning and representation framework formulated by Ryszard S. Michalski, integrating aspects of predicate logic, propositional calculus, and multi-valued logic. Attributional calculus serves as a structured language for natural induction, an inductive learning approach that yields results in a format intuitive to humans.

‍

Augmented Reality (AR)

An immersive interaction with a real-world setting where elements from the physical environment are enhanced by digitally generated perceptual information, often spanning multiple sensory channels such as visual, auditory, haptic, somatosensory, and olfactory stimuli.

‍

Auto-Classification

The use of a machine learning system, natural language processing (NLP), and other AI methods to automate text classification with greater speed, efficiency, and accuracy.

‍

Auto-Complete

Auto-complete is a search feature that proposes potential queries based on the text being entered to formulate a search request.

‍

Autoencoder

A category of artificial neural networks designed to acquire optimized representations of unlabelled data through unsupervised learning. A widely used implementation of this approach is the variational autoencoder (VAE).

‍

Automata Theory

The exploration of theoretical machines and automata, along with the computational challenges that can be addressed using them. This field is a branch of theoretical computer science and discrete mathematics.

‍

Automated Machine Learning (AutoML)

A branch of machine learning (ML) focused on the automatic optimization of an ML system to increase its effectiveness.

‍

Automated Planning and Scheduling

A subdivision of artificial intelligence focused on devising strategies or action sequences, usually intended for implementation by intelligent agents, self-governing robots, and autonomous vehicles. Unlike conventional control and classification challenges, solutions are intricate and must be identified and refined within a multidimensional space. Planning is also linked to decision theory.

‍

Automated Reasoning

A domain within computer science and mathematical logic focused on analyzing various facets of reasoning. Research in automated reasoning facilitates the development of software that helps computers infer conclusions fully or almost entirely autonomously. While automated reasoning is regarded as a subset of AI, it also intersects with theoretical computer science and even philosophy.

‍

Autonomic Computing (AC)

The self-regulating attributes of decentralized computing resources that adjust to unforeseen variations while concealing underlying intricacies from administrators and end-users. Introduced by IBM in 2001, this initiative was ultimately intended to create computing systems with autonomous management capabilities, addressing the escalating complexity of system administration and mitigating the obstacles that this complexity presents to continued advancement.

‍

Autonomous Car

A transport system capable of perceiving its surroundings and navigating with minimal or no human intervention.

‍

Autonomous Robot

A machine that executes actions or functions with a significant level of independence. Autonomous robotics is generally regarded as a branch of AI, robotics, and information engineering.

‍

Backpropagation

A technique utilized in artificial neural networks to determine a gradient essential for computing the weights applied within the network. Backpropagation is an abbreviation for backward propagation of errors, as the error is measured at the output and propagated in reverse through the network's layers. It is frequently employed to train deep neural networks, which are neural architectures containing multiple hidden layers.

‍

Backpropagation Through Structure (BPTS)

A gradient-based method for optimizing recurrent neural networks, introduced in a 1996 research paper authored by Christoph Goller and Andreas Küchler.

‍

Backpropagation Through Time (BPTT)

Reword the following using synonyms wherever possible but without changing the meaning:

‍

Backward Chaining

A reasoning technique commonly referred to as operating in reverse from the desired outcome. It is employed in automated proof systems, deduction engines, verification assistants, and various artificial intelligence applications.

‍

Bag-of-Words Model

A streamlined representation utilized in natural language processing and information retrieval (IR). Within this framework, a text (such as a sentence or document) is depicted as a collection, or multiset, of its words, ignoring syntax and word sequence while preserving word frequency. The bag-of-words model has also been applied in computer vision. It is frequently employed in document categorization techniques, where the presence and frequency of each word serve as attributes for training a classifier.

‍

Bag-of-Words Model in Computer Vision

In the field of computer vision, the bag-of-words (BoW) approach can be utilized for image categorization by interpreting image attributes as words. In text classification, a bag of words represents a sparse vector containing word occurrence frequencies, which is essentially a sparse histogram over a defined lexicon. Similarly, in computer vision, a bag of visual words consists of a vector quantifying the frequency of a predefined set of local image descriptors.

‍

Batch Normalization

A method for enhancing the efficiency and robustness of ANNs. It is a strategy to supply any layer in a neural network with inputs that have zero mean and unit variance. Batch normalization was first presented in a 2015 research paper. It functions by standardizing the input layer through modification and rescaling of activations.

‍

Bayesian Programming

A framework and an approach for establishing a method to define probabilistic models and address problems when incomplete information is present.

‍

Bees Algorithm

A population-based optimization technique introduced by Pham, Ghanbarzadeh, and collaborators in 2005. It emulates the food-seeking behavior of honeybee swarms. In its fundamental form, the algorithm integrates localized exploration with broad search strategies and is applicable to both combinatorial and continuous optimization problems. The sole prerequisite for employing the bees algorithm is the definition of a distance metric between potential solutions. Numerous studies have validated its efficacy and unique capabilities.

‍

Behavior Informatics (BI)

The analysis of behavioral data to extract intelligence and insights regarding actions and tendencies.

‍

Behavior Tree (BT)

‍

A mathematical framework for executing plans, widely applied in computer science, robotics, control systems, and video games. It models transitions between a finite collection of tasks in a structured manner. Its strength lies in its capability to construct highly intricate behaviors from simpler tasks without concern for their underlying implementation. Behavior Trees (BTs) share similarities with hierarchical state machines, with the key distinction that their fundamental unit is a task rather than a state. Their intuitive nature makes them less error-prone and highly favored among game developers. BTs have been demonstrated to generalize multiple other control architectures.

‍

Belief–Desire–Intention Software Model (BDI)

A computational framework designed for developing intelligent agents. While it is superficially recognized for incorporating an agent's beliefs, desires, and intentions, it fundamentally employs these notions to address specific challenges in agent-based programming. At its core, it establishes a mechanism for distinguishing between the selection of a plan, either from a predefined plan library or an external planning system, and the execution of currently active plans. As a result, BDI agents can effectively balance the time allocated to deliberating over plans and carrying them out. However, the process of generating plans falls outside the model's scope and is left to the system designer and developer.

‍

Bias-Variance Tradeoff

In statistics and machine learning, the bias-variance tradeoff refers to the characteristic of a collection of predictive models in which models exhibiting lower bias in parameter estimation tend to have greater variability in parameter estimates across different samples, and conversely, models with higher bias demonstrate reduced variance.

‍

Bidirectional Coder Representation From Transformers (BERT)

A large-scale pre-trained model initially trained on vast amounts of unlabelled data. It is then adapted to a specific NLP task by being provided with a smaller, task-focused dataset for fine-tuning the final model.

‍

Big Data

Big data refers to extensive data sets that can be analyzed to uncover patterns and trends that inform business decisions. It is termed big data because organizations can now accumulate vast and intricate data using various collection tools and systems. Big data can be gathered rapidly and stored in multiple formats.

‍

Big O Notation

A symbolic representation that characterizes the asymptotic behavior of a function as its input approaches a specific value or extends to infinity. It belongs to a group of notations introduced by Paul Bachmann, Edmund Landau, and others, collectively referred to as Bachmann–Landau notation or asymptotic notation.

‍

Binary Tree

A hierarchical data structure where each element, known as a node, has at most two subordinate nodes, termed the left child and the right child. A recursive characterization using set theory defines a non-empty binary tree as an ordered triple (L, S, R), where L and R represent either binary trees or an empty set, and S is a singleton set. Some scholars also permit a binary tree to be entirely empty.

‍

Blackboard System

An artificial intelligence methodology rooted in the blackboard architectural paradigm, wherein a shared information repository, known as the blackboard, is progressively modified by a diverse set of expert knowledge modules. The process begins with a problem definition and concludes with a resolved outcome. Each knowledge module contributes to the blackboard by appending a partial solution whenever its internal conditions align with the current state of the blackboard. Through this collaborative mechanism, the specialists collectively work toward problem resolution.

‍

Black Boxes

We refer to things we do not comprehend as black boxes because their inner workings are hidden from view. Many machine learning models are considered black boxes, meaning we lack insight into how they utilize different aspects of the data when making decisions. Although we typically know which features are used, we do not fully understand how they are applied. Presently, there are two main approaches to uncovering the inner workings of AI models: interpretable machine learning and explainable machine learning.

‍

Boltzmann Machine

A category of probabilistic recurrent neural network and Markov random field. Boltzmann machines can be regarded as the stochastic, generative equivalent of Hopfield networks.

‍

Boolean Satisfiability Problem

The challenge of determining whether there exists an assignment of values that satisfies a specified Boolean expression. In other words, it investigates whether the variables within a given Boolean formula can be consistently substituted with TRUE or FALSE in such a manner that the expression evaluates to TRUE. If such an assignment is possible, the formula is termed satisfiable. Conversely, if no such assignment exists, meaning the function represented by the formula remains FALSE for all possible variable configurations, the formula is deemed unsatisfiable. For instance, the expression "x AND NOT y" is satisfiable because assigning x = TRUE and y = FALSE results in (x AND NOT y) = TRUE. In contrast, "x AND NOT y" is unsatisfiable.

‍

Boosting

A machine learning ensemble optimization technique primarily aimed at minimizing bias rather than variance by training models in a sequential manner, with each successive model adjusting for the errors made by its predecessor.

‍

Bootstrap Aggregating

A machine learning ensemble optimization method primarily designed to decrease variance, rather than bias, by training multiple models separately and combining their predictions through averaging.

‍

Brain Technology

A technology that leverages the most recent advancements in neuroscience. The term was originally coined by the Artificial Intelligence Laboratory in Zurich, Switzerland, within the scope of the ROBOY project. Brain technology can be utilized in robotics, knowledge management systems, and various other applications with self-learning abilities. Specifically, Brain Technology implementations support the visualization of the underlying learning framework, frequently referred to as know-how maps.

‍

Branching Factor

In computing, hierarchical data structures, and game theory, the quantity of descendant nodes at each parent node, known as the outdegree. If this number varies by node, an average branching factor can be determined.

‍

Brute-Force Search

A highly versatile problem-solving method and algorithmic framework that involves methodically listing all potential solution candidates and verifying whether each one meets the conditions specified by the problem.

Capsule Neural Network (CapsNet)

A machine learning framework that represents a category of ANNs designed to more effectively capture hierarchical relationships. This methodology seeks to emulate the structural organization of biological neural systems more accurately.

‍

Case-Based Reasoning (CBR)

Generally interpreted, the method of resolving novel problems by leveraging solutions from previously encountered, analogous problems.

‍

Cataphora

In linguistics, a cataphora is a reference that appears before the noun it refers to. For example, in the sentence, "Though he enjoyed the entrée, Paul didn't like the appetizers," the word "he" functions as a cataphora.

‍

Categorization

Classification is a natural language processing capability that assigns a specific category to a document.

‍

Category

A category is a designation given to a document to characterize the information it contains.

‍

Category Trees

Allows you to see all the rule-based classifications within a collection. It is used to create, remove, and modify the rules that link documents to categories. Also known as a taxonomy, it is structured in a hierarchical format.

‍

Chat-Based Generative Pre-Trained Transformer (ChatGPT) Models

A system constructed with a neural network transformer-style AI model that excels in natural language processing tasks. In this scenario, the model: 1. can produce replies to inquiries (Generative); 2. was pre-trained on a substantial amount of textual data from the internet (Pre-trained); 3. and can analyze sentences in a way that differs from other model types (Transformer).

‍

Chatbot

A chatbot is a software program created to simulate human dialogue using text or voice inputs.

‍

Classification

Methods that allocate a set of predefined labels to unstructured text, allowing for the organization, structuring, and classification of various types of text, ranging from documents and medical records to emails and files, within applications and on websites or social media platforms.

‍

Cloud Robotics

A branch of robotics that seeks to integrate cloud-based technologies, including cloud computing, cloud storage, and various Internet-based solutions, to harness the advantages of converged infrastructure and shared services for robotic applications. By connecting to the cloud, robots can utilize the vast computational power, storage capacity, and communication capabilities of modern cloud data centers, allowing them to process and exchange information with other robots, intelligent systems, or human users. Additionally, humans can assign tasks to robots remotely via networked connections. Cloud computing allows robotic systems to achieve expanded functionality while minimizing costs, facilitating the development of lightweight, cost-efficient, and more intelligent robots with a "brain" hosted in the cloud. This cloud-based intelligence encompasses data centers, knowledge repositories, task planning modules, deep learning models, information processing units, environmental models, and communication frameworks.

‍

Cluster Analysis

The process of organizing a collection of items into subsets, known as clusters, such that elements within the same cluster exhibit greater similarity, according to a specific criterion, compared to those in different clusters. This serves as a fundamental operation in exploratory data mining and is a widely employed method in statistical data analysis. Clustering finds applications in many fields, including machine learning, pattern recognition, image processing, information retrieval, bioinformatics, data compression, and computer graphics.

‍

COBWEB

A progressive framework for hierarchical conceptual grouping. COBWEB was devised by Professor Douglas H. Fisher, presently at Vanderbilt University. This method incrementally structures observations into a taxonomy tree, where each node signifies a category and is characterized by a probabilistic representation that encapsulates the attribute-value distributions of entities classified under it. The resulting classification tree can be utilized to infer missing attributes or determine the category of a novel entity.

‍

Cognitive Architecture

The Institute of Creative Technologies characterizes cognitive architecture as a theoretical framework regarding the immutable structures that constitute a mind, whether in biological or synthetic systems, and the manner in which they interact alongside the knowledge and competencies embedded within the framework to produce intelligent conduct in a variety of intricate settings.

‍

Cognitive Computing

Cognitive computing is essentially synonymous with artificial intelligence. It is a computational model designed to replicate human cognitive processes, including pattern recognition and learning. Marketing teams sometimes use this term to remove the science-fiction connotations associated with AI.

‍

Cognitive Map

A cognitive map, also referred to as a mental palace, is an internal representation that helps an individual absorb, encode, retain, retrieve, and interpret information regarding the spatial relationships and characteristics of elements within their surroundings.

‍

Cognitive Science

The cross-disciplinary scientific investigation of cognition and its mechanisms.

‍

Combinatorial Optimization

In operations research, applied mathematics, and theoretical computer science, combinatorial optimization is a field focused on identifying the best possible element from a limited collection of choices.

‍

Committee Machine

A category of artificial neural network that employs a divide-and-conquer approach, where the outputs of multiple neural networks are integrated into a unified response. The aggregated output of the committee machine is intended to surpass the performance of its individual expert components. Contrast with classifier ensembles.

‍

Commonsense Knowledge

In artificial intelligence studies, commonsense knowledge comprises universally known facts about the everyday world, such as "Limes have a sour taste," which all people are presumed to understand. The earliest AI system designed to handle commonsense knowledge was Advice Taker, developed by John McCarthy in 1959.

‍

Commonsense Reasoning

A subdivision of AI focused on replicating the human capacity to infer assumptions regarding the nature and characteristics of commonplace situations they experience daily.

‍

Completions

The result produced in response to a generative input.

‍

Composite AI

The integrated use of various AI methods to increase learning efficiency, expand the scope of knowledge representations, and ultimately address a broader array of business challenges more effectively.

‍

Computational Chemistry

A field of chemistry that employs computational modeling to aid in resolving chemical challenges.

‍

Computational Complexity Theory

Centers on categorizing computational challenges based on their intrinsic complexity and establishing relationships among these categories. A computational challenge refers to a task executed by a computer. Such a problem can be resolved through the systematic application of mathematical procedures, such as an algorithm.

‍

Computational Creativity

A cross-disciplinary pursuit encompassing the domains of AI, cognitive psychology, philosophy, and the creative arts.

‍

Computational Cybernetics

The fusion of cybernetics with computational intelligence methodologies.

‍

Computational Humor

A subfield of computational linguistics and artificial intelligence that leverages computers for the study of humor.

‍

Computational Intelligence (CI)

Typically denotes a computer's capability to acquire knowledge of a particular task through data analysis or empirical observation.

‍

Computational Learning Theory

In computing, computational learning theory, often abbreviated as learning theory, is a branch of AI focused on examining the development and evaluation of machine learning algorithms.

‍

Computational Linguistics

Computational linguistics is a multidisciplinary domain focused on the algorithmic representation and processing of natural language.

‍

Computational Mathematics

The mathematical investigation in scientific domains where computation serves a fundamental role.

‍

Computational Neuroscience

A field of neuroscience that utilizes mathematical frameworks, theoretical examination, and abstractions of the brain to comprehend the fundamental principles guiding the formation, organization, functionality, and cognitive capabilities of the nervous system.

‍

Computational Number Theory

The exploration of computational methods for executing arithmetic operations within number theory.

‍

Computational Problem

In theoretical computing, a computational problem is a mathematical construct that encapsulates a set of queries that computers may be capable of resolving.

‍

Computational Semantics

Computational semantics involves the exploration of methods to automate the creation and interpretation of meaning representations for natural language expressions.

‍

Computational Statistics

The juncture between computational science and statistical analysis.

‍

Computer-Automated Design (CAutoD)

Design automation typically refers to electronic design automation (EDA) or Design Automation as a Product Configurator. Expanding upon Computer-Aided Design (CAD), automated design and computer-automated design encompass a wider array of applications, including automotive engineering, civil engineering, composite material development, control systems engineering, dynamic system identification and optimization, financial modeling, industrial machinery, mechatronic systems, steel structures, structural optimization, and the creation of innovative systems. More recently, conventional CAD simulation has been evolving into CAutoD through biologically inspired machine learning, incorporating heuristic search strategies such as evolutionary computation and swarm intelligence algorithms.

‍

Computer Science

The principles, practical investigation, and applied engineering that underpin the conception and utilization of computing devices. It involves developing algorithms that manipulate, retain, and transmit digital data. A computing scientist focuses on the fundamentals of computation and the architecture of computational frameworks.

‍

Computer Vision

Computer vision is a multidisciplinary domain within science and technology that concentrates on helping computers interpret and extract insights from images and videos. For AI developers, computer vision facilitates the automation of tasks traditionally carried out by the human visual system.

‍

Concept Drift

In prognostic analytics and machine learning, concept drift refers to the phenomenon where the statistical characteristics of the target variable, which the model aims to forecast, evolve unpredictably. This presents challenges as the accuracy of predictions deteriorates progressively with time.

‍

Connectionism

A methodology within cognitive science that seeks to elucidate mental phenomena through the utilization of artificial neural networks.

‍

Consistent Heuristic

In the exploration of pathfinding challenges within artificial intelligence, a heuristic function is considered consistent or monotonic if its estimation never exceeds the projected distance from any adjacent vertex to the destination, combined with the cost required to reach that neighbor.

‍

Constrained Conditional Model (CCM)

A machine learning and reasoning framework that improves the training of conditional models by incorporating declarative constraints.

‍

Constraint Logic Programming

A variant of constraint programming where logic programming is expanded to incorporate principles from constraint satisfaction. A constraint logic program consists of a logic-based framework that integrates constraints within clause bodies. For instance, in the clause A(X,Y) :- X+Y>0, B(X), C(Y), the expression X+Y>0 represents a constraint, while A(X,Y), B(X), and C(Y) are literals, as in conventional logic programming. This clause specifies a condition under which A(X,Y) is valid: X+Y must be greater than zero, and both B(X) and C(Y) must hold true.

‍

Constraint Programming

A programming model in which relationships among variables are expressed as constraints. Unlike the standard constructs of imperative programming languages, constraints do not define a sequence of operations to perform but instead describe the characteristics that a valid solution must satisfy.

‍

Constructed Language

A linguistic system whose sound patterns, syntax, and lexicon are intentionally created rather than evolving organically. Constructed languages are also known as artificial, designed, or fabricated languages.

‍

Content

Distinct units of information, namely, documents, that can be aggregated to create training datasets or produced by generative AI.

‍

Content Enrichment

The method of utilizing sophisticated approaches like machine learning, AI, and natural language processing to automatically derive valuable insights from text-based documents.

‍

Controlled Vocabulary

A controlled vocabulary is a carefully selected set of terms and expressions pertinent to a particular application or industry. These elements may include additional attributes that define their linguistic behavior and the meanings they convey in relation to topics and other contexts.

Although the purpose of a managed vocabulary is comparable to that of a taxonomy, they differ in that taxonomy nodes serve as category labels, whereas the nodes in a managed vocabulary represent specific words and phrases that must be identified within a text.

‍

Control Theory

In the domain of control systems engineering, a specialized branch of mathematics focuses on regulating continuously functioning dynamic systems within designed processes and machinery. The goal is to formulate a control framework that governs these systems through an optimal control response, minimizing delay and overshoot while maintaining stability.

‍

Conversational AI

Utilized by developers to create interactive user experiences, chatbots, and virtual assistants for many applications. These platforms support integration with communication channels such as messaging apps, social networks, SMS, and websites. A conversational AI platform includes a developer API, allowing third parties to tailor the system with their own custom modifications.

‍

Convolutional Neural Networks (CNN)

A category of deep learning neural networks comprising one or multiple layers, designed for image analysis and recognition.

‍

Co-Occurrence

Co-occurrence typically denotes the appearance of distinct elements within the same document. It is frequently applied in business intelligence to intuitively identify patterns and infer relationships between concepts that are not inherently linked. For example, if an investor's name frequently appears in articles about startups securing funding, it could be interpreted that the investor has a strong ability to select successful investments.

‍

Corpus

The complete collection of linguistic data intended for analysis. More precisely, a corpus is a well-structured compilation of documents that should accurately reflect the types of texts an NLP system will encounter in real-world use, both in terms of subject matter and the distribution of themes and ideas.

‍

Critical AI

Critical AI is a method of analyzing artificial intelligence through a lens that emphasizes thoughtful evaluation and critique to comprehend and question both current and past frameworks within AI. Learn more about Critical AI.

‍

Crossover

In genetic algorithms and evolutionary computation, a genetic mechanism employed to merge the hereditary data of two parent solutions to produce novel offspring. It serves as a stochastic method for deriving new solutions from a pre-existing population, resembling the recombination process occurring in sexual reproduction among biological entities. Additionally, solutions can be produced through direct duplication of an existing solution, akin to asexual reproduction. Newly derived solutions are generally subjected to mutation before being incorporated into the population.

‍

Custom/Domain Language Model

A model designed explicitly for a particular organization or sector, such as the insurance industry.

Darkforest

A computer Go application created by Facebook, utilizing deep learning methodologies with a convolutional neural network. Its improved iteration, Darkforest2, integrates the strategies of its predecessor with Monte Carlo tree search (MCTS). The MCTS efficiently adapts tree search techniques, frequently employed in computer chess programs, by introducing stochastic elements. Following this upgrade, the system is referred to as Darkfmcts3.

‍

Dartmouth Workshop

The Dartmouth Summer Research Project on Artificial Intelligence was the title of a 1956 summer workshop, now widely regarded, though not universally, as the foundational milestone for AI as a discipline.

‍

Data Augmentation

Data augmentation in data analysis refers to methodologies employed to expand the volume of data. It aids in minimizing overfitting when training a machine learning model.

‍

Data Discovery

The process of discovering valuable data insights and delivering them to the appropriate users at the right time.

‍

Data Drift

Data drift takes place when the distribution of input data evolves, also referred to as covariate shift.

‍

Data Extraction

Data extraction is the method of gathering or obtaining various types of raw data from multiple sources, many of which may be disorganized or entirely unstructured.

‍

Data Fusion

The procedure of merging various data sources to generate more coherent, precise, and valuable information than that offered by any single data source. 

‍

Data Ingestion

The method of acquiring diverse data from various sources, reorganizing it, and converting it into a standardized format or repository to increase usability.

‍

Data Integration

The method of merging data stored in distinct sources and presenting users with a consolidated perspective of them. This procedure gains importance in numerous scenarios, including business-related ones, such as when two comparable firms must unify their databases, and scientific fields. Data integration emerges more frequently as the scale of data expands and the necessity to share existing information surges. It has become the subject of substantial theoretical investigation, with many unresolved challenges still remaining.

‍

Data Labeling

A method of labeling data to help machines identify objects. Additional information is applied to different data formats, such as text, audio, images, and video, to generate metadata used for training AI models.

‍

Datalog

A declarative logic-based programming language that, in terms of syntax, forms a subset of Prolog. It is frequently employed as a query language for deductive database systems. In recent times, Datalog has gained renewed utility in domains such as data fusion, knowledge extraction, networking, software analysis, cybersecurity, and cloud-based computing.

‍

Data Mining

Data mining is the practice of analyzing extensive data sets to uncover patterns that improve models or address challenges.

‍

Data Scarcity

The absence of sufficient data that could potentially fulfill the system's requirement to increase the precision of predictive analytics.

‍

Data Science

Data science is a multidisciplinary field of technology that leverages algorithms and methodologies to collect and examine vast amounts of data, revealing patterns and insights that guide business decisions.

‍

Data Set

An aggregation of information. Typically, a dataset corresponds to the contents of a singular database table or a single statistical data matrix, where each column in the table signifies a specific variable, and every row represents a distinct entity within the dataset. The dataset enumerates values for each variable, such as an object's height and weight, for every included entity. Each individual value is referred to as a datum. The dataset may encompass information for one or multiple entities, corresponding to the total number of rows.

‍

Data Warehouse

A framework employed for reporting and data examination. Data warehouses serve as centralized repositories that consolidate information from multiple distinct sources. They retain both present and past data within a unified location.

‍

Decision Boundary

For ANNs or perceptrons utilizing backpropagation, the nature of the decision boundary that the network can model is influenced by the quantity of hidden layers. Without hidden layers, the network is restricted to learning only linear problems. With a single hidden layer, it can approximate any continuous function over compact subsets of Rⁿ, as established by the Universal Approximation Theorem, enabling it to represent an arbitrary decision boundary.

‍

Decision Support System (DSS)

An information system designed to aid business or organizational decision-making processes. Decision support systems (DSSs) assist management, operational, and strategic planning levels within an organization, typically mid-to-upper management, by facilitating decision-making for issues that evolve quickly and are not easily predefined, such as unstructured or semi-structured decision challenges. These systems can be entirely automated, human-controlled, or a hybrid of both approaches.

‍

Decision Theory

The examination of the logic behind an agent's decision-making process. Decision theory is divided into two main areas: normative decision theory, which provides guidance on making optimal choices based on a given set of uncertain assumptions and preferences, and descriptive decision theory, which investigates how real-world agents, potentially exhibiting irrational behavior, actually arrive at decisions.

‍

Decision Tree Learning

Employs a decision tree as a forecasting framework to map observations regarding an entity, depicted in the branches, to determine the entity's target attribute represented in the leaves. This technique is among the predictive modeling methodologies utilized in statistics, data mining, and machine learning.

‍

Declarative Programming

A coding paradigm that conveys the logic of a computation without specifying its execution sequence.

‍

Deductive Classifier

A category of artificial intelligence reasoning engine. It processes as input a collection of statements in a frame-based language concerning a domain like medical studies or molecular biology. This includes, for instance, the names of categories, subcategories, attributes, and constraints on permissible values.

‍

Deep Blue

Deep Blue was a computer system designed by IBM for playing chess. It is recognized as the first chess-playing machine to secure both a game and an entire match victory against a reigning world champion under standard time constraints.

‍

Deep Learning

Deep learning is a branch of AI that mimics the human brain's structure by analyzing how it organizes and processes information to make decisions. Rather than depending on an algorithm designed for a single specific task, this subset of machine learning can extract insights from unstructured data without human oversight.

‍

DeepMind Technologies

A British AI firm established in September 2010, now a subsidiary of Alphabet Inc. Headquartered in London, it also operates research centers in Canada, France, and the United States. Acquired by Google in 2014, the company developed a neural network capable of learning to play video games in a manner akin to human learning, as well as a neural Turing machine, which is a neural network designed to interface with external memory similarly to a traditional Turing machine, enabling it to replicate aspects of human short-term memory. The company gained global attention in 2016 when its AlphaGo system triumphed over professional Go player Lee Sedol, the reigning world champion, in a five-game series, an event documented in a film. A more advanced system, AlphaZero, surpassed the strongest existing programs in Go, chess, and shogi, or Japanese chess, after only a few days of self-play using reinforcement learning.

‍

Default Logic

A non-monotonic reasoning framework introduced by Raymond Reiter to systematize inference based on default presumptions.

‍

Density-Based Spatial Clustering of Applications With Noise (DBSCAN)

A grouping algorithm introduced by Martin Ester, Hans-Peter Kriegel, Jörg Sander, and Xiaowei Xu in 1996.

‍

Description Logic (DL)

A collection of structured knowledge representation languages. Numerous description logics (DLs) offer greater expressiveness than propositional logic while remaining less expressive than first-order logic. Unlike the latter, fundamental reasoning tasks in DLs are typically decidable, and efficient computational procedures have been developed and implemented to address these tasks. There exist general, spatial, temporal, spatiotemporal, and fuzzy variants of description logics, each striking a unique balance between expressiveness and computational complexity by incorporating different sets of mathematical constructs.

‍

Developmental Robotics (DevRob)

A scientific discipline focused on investigating the developmental processes, frameworks, and limitations that support continuous and unrestricted acquisition of new abilities and knowledge in physically embedded machines.

‍

Diagnosis

Focused on creating algorithms and methodologies capable of verifying whether a system's behavior is accurate. If the system operates incorrectly, the algorithm should identify, with maximum precision, the malfunctioning component and the nature of the fault. The analysis relies on observations that supply insights into the system's present functioning.

‍

Dialogue System

A computational system designed to engage in structured interactions with a human. Conversational systems have utilized text, speech, visuals, touch-based inputs, gestures, and various other modalities for communication on both input and output channels.

‍

Did You Mean (DYM)

Did You Mean is a natural language processing feature utilized in search applications to detect misspellings in a query or propose alternative queries that may yield relevant results within the search database.

‍

Diffusion Model

In machine learning, diffusion models, also referred to as diffusion probabilistic models or score-based generative models, constitute a category of latent variable models. These models function as Markov chains trained via variational inference. The objective of diffusion models is to capture the latent structure of a dataset by simulating how data points propagate through the latent space. In the realm of computer vision, this entails training a neural network to restore images degraded by Gaussian noise by learning to reverse the diffusion process. Diffusion models primarily comprise three key components: the forward process, the reverse process, and the sampling mechanism. Notable diffusion modeling frameworks utilized in computer vision include denoising diffusion probabilistic models, noise-conditioned score networks, and stochastic differential equations.

‍

Dijkstra's Algorithm

An algorithm designed to determine the shortest routes between vertices in a weighted graph, which can symbolize, for instance, transportation networks.

‍

Dimensionality Reduction

The procedure of minimizing the quantity of stochastic variables being analyzed by deriving a set of key variables. It can be categorized into feature selection and feature extraction.

‍

Disambiguation

Disambiguation, also known as word-sense disambiguation, is the method of eliminating ambiguity in terms that have multiple meanings, preventing misinterpretation of the same sequence of text.

‍

Discrete System

Any mechanism possessing a countable set of states. Discrete systems can be contrasted with continuous systems, also referred to as analog systems. A finite discrete system is frequently represented using a directed graph and examined for accuracy and complexity within computational theory. Since discrete systems have a countable number of states, they can be precisely described through mathematical models. A computer functions as a finite-state machine and can be considered a discrete system. As computers are commonly employed to simulate both discrete and continuous systems, techniques have been devised to represent real-world continuous systems in a discrete manner. One such approach entails sampling a continuous signal at distinct time intervals.

‍

Distributed Artificial Intelligence (DAI)

A branch of artificial intelligence research focused on creating decentralized solutions for various problems. Distributed AI is closely associated with and serves as a forerunner to the 

‍

Domain Knowledge

The knowledge and proficiency your organization has accumulated.

‍

Double Descent

A statistical and machine learning phenomenon in which both a model with few parameters and one with an exceptionally high number of parameters exhibit low test error, whereas a model with a parameter count roughly equivalent to the number of training data points incurs high error. This occurrence has been viewed as unexpected, as it challenges traditional assumptions regarding overfitting in classical machine learning.

‍

Dropout

A normalization method for minimizing overfitting in ANNs by inhibiting intricate co-adaptations on training datasets.

‍

Dynamic Epistemic Logic (DEL)

A formal system addressing the dynamics of knowledge and information modification. Generally, DEL concentrates on scenarios with multiple agents and examines how their awareness evolves in response to occurring events.

Eager Learning

A training approach where the system endeavors to formulate a broad, input-agnostic target function during the learning phase, in contrast to lazy learning, where the process of generalization beyond the training dataset is postponed until the system receives a query.

‍

Early Stopping

A constraint enforcement method frequently employed during the training of a machine learning model using an iterative approach like gradient descent.

‍

Ebert Test

An assessment designed to measure whether a computer-generated synthetic voice can deliver a joke proficiently enough to make listeners laugh. Film critic Roger Ebert introduced this concept at the 2011 TED conference as a challenge for software engineers to develop an artificial voice capable of replicating the nuances of human speech, including intonation, timing, delivery, and inflection. This evaluation is analogous to the Turing test, proposed by Alan Turing in 1950, which assesses a machine's ability to demonstrate intelligent behavior indistinguishable from that of a human.

‍

Echo State Network (ESN)

A recurrent neural network featuring a sparsely interconnected hidden layer, typically with around 1% connectivity. The connections and weights of the hidden neurons are predetermined and assigned at random. The weights of the output neurons, however, can be adjusted to help the network generate or replicate specific temporal patterns. The primary advantage of this network lies in the fact that, despite exhibiting nonlinear behavior, only the weights linking the hidden neurons to the output neurons are updated during training. Consequently, the error function remains quadratic concerning the parameter vector, allowing for straightforward differentiation into a linear system.

‍

Edge Model

A model that incorporates data usually located outside centralized cloud data centers and nearer to local devices or users, such as wearables and Internet of Things (IoT) sensors or controllers.

‍

Embedding

A collection of data frameworks within a large language model (LLM) that represents a body of text, where words are encoded as high-dimensional vectors. This approach increases the efficiency of processing meaning, translation, and the creation of new content.

‍

Embodied Agent

A cognitive agent that engages with its surroundings using a tangible body within that environment. Agents depicted visually with a form, such as a humanoid figure or an animated creature, are also referred to as embodied agents, even though their embodiment exists solely in a virtual rather than a physical form.

‍

Embodied Cognitive Science

A cross-disciplinary area of study focused on elucidating the mechanisms that drive intelligent conduct. It encompasses three primary approaches: 1) the simulation of cognitive and biological systems in an integrative way that treats the mind and body as a unified whole, 2) the establishment of a universal framework of fundamental principles governing intelligent behavior, and 3) the empirical deployment of robotic entities within regulated settings for experimentation.

‍

Emergent Behavior

Emergent behavior, also known as emergence, occurs when an AI system exhibits unexpected or unplanned abilities.

‍

Emotion AI or Affective Computing

AI that evaluates a user's emotional condition through computer vision, voice/audio input, sensors, or software algorithms. It can trigger responses by executing tailored actions to align with the user's mood.

‍

Ensemble Learning

The utilization of multiple machine learning models to achieve superior predictive accuracy compared to what any individual underlying algorithm could accomplish independently.

‍

Entity

An entity is any noun, term, or expression within a document that denotes a concept, individual, object, or abstraction. This category also encompasses quantifiable elements.

‍

Environmental, Social, and Governance (ESG)

Originally associated with business and government, relating to an organization's social influence and responsibility, disclosures in this domain are regulated by a combination of mandatory and voluntary compliance frameworks.

‍

Epoch

In machine learning, especially in the development of ANNs, an epoch refers to a single complete pass through the entire training dataset during model training. Smaller models are generally trained for multiple epochs until optimal performance is achieved on the validation dataset, whereas the largest models may undergo training for just one epoch.

‍

Error-Driven Learning

A branch of machine learning focused on determining how an agent should execute actions within an environment to reduce a specified error signal. It represents a form of reinforcement learning.

‍

Ethics of Artificial Intelligence

The branch of technology ethics that pertains specifically to artificial intelligence.

‍

ETL or Extract, Transform, Load

Entity extraction is a natural language processing capability that detects and identifies significant entities within a document.

‍

Evolutionary Algorithm (EA)

A branch of evolutionary computation, a broad population-based metaheuristic optimization method. An evolutionary algorithm employs principles derived from biological evolution, including reproduction, mutation, recombination, and selection. Potential solutions to the optimization problem function as individuals within a population, while a fitness function evaluates their quality, similar to a loss function. The population evolves through the iterative application of these operators.

‍

Evolutionary Computation

A class of algorithms for worldwide optimization influenced by natural evolution, as well as the domain of artificial intelligence and soft computing that examines these methods. In precise terms, they constitute a group of population-based, trial-and-error problem-solving techniques with a metaheuristic or probabilistic optimization nature.

‍

Evolving Classification Function (ECF)

Adaptive classification functions are utilized for categorization and grouping within machine learning and AI, commonly applied in data stream analysis tasks within fluid and evolving environments.

‍

Existential Risk

The conjecture that significant advancements in artificial general intelligence (AGI) might eventually lead to the extinction of humanity or another irreversible worldwide disaster.

‍

Expert System

A computational system that replicates the decision-making capabilities of a human specialist. Expert systems are engineered to tackle intricate problems by logically processing extensive knowledge bases, primarily represented as if-then rules rather than traditional procedural programming.

‍

Explainable AI/Explainability

An AI methodology in which the functioning of its algorithms is transparent and comprehensible to humans. Unlike black-box AI, this approach provides visibility into the decision-making process and the rationale behind its outcomes.

‍

Extraction or Keyphrase Extraction

Several terms that capture the core concepts and fundamental meaning of the text within documents.

‍

Extractive Summarization

Extracts key details from a text and clusters related fragments to create a succinct summary.

Fast-And-Frugal Trees

A form of decision tree used for classification. Fast-and-frugal trees serve as decision-making instruments that function as lexicographic classifiers and, when necessary, assign an action or choice to each category or class.

‍

Feature

A distinct quantifiable attribute or trait of a phenomenon. In computer vision and image analysis, a feature represents a fragment of data regarding an image's content, usually indicating whether a specific area of the image possesses particular characteristics. Features can encompass identifiable structures within an image, such as key points, contours, or objects, or emerge from a broader regional computation or feature extraction process applied to the image.

‍

Feature Extraction

In machine learning, pattern recognition, and image analysis, feature extraction begins with an original collection of recorded data and generates transformed values, or features, designed to be insightful and non-redundant. This process aids in the subsequent phases of learning and generalization and, in certain instances, supports human interpretability.

‍

Feature Learning

In machine learning, feature learning, also known as representation learning, encompasses a collection of methods enabling a system to autonomously identify the representations essential for feature recognition or categorization from unprocessed data. It eliminates the need for manual feature engineering, allowing the machine to simultaneously acquire the features and apply them to execute a particular task.

‍

Feature Selection

In machine learning and statistics, feature selection, also referred to as variable selection, attribute selection, or variable subset selection, is the procedure of identifying a subset of pertinent features, such as variables and predictors, for incorporation in model development.

‍

Federated Learning

A machine learning approach that supports the training of models on multiple devices using distributed data, thereby aiding in safeguarding the privacy of individual users and their information.

‍

Few-Shot Learning

Unlike conventional models that rely on extensive training datasets, few-shot learning requires only a limited number of training examples to generalize effectively and generate meaningful results.

‍

Fine-Tuned Model

A model tailored to a particular domain or classification of information, such as a subject, sector, or set of challenges.

‍

Fine-Tuning

Enhancing a previously trained model by further refining it with new data tailored to a specific context or task.

‍

First-Order Logic

A set of formal frameworks utilized in mathematics, philosophy, linguistics, and computer science. First-order logic employs quantified variables over non-logical entities and permits the formation of statements containing variables. Instead of fixed propositions like "Aristotle is a man," it allows expressions such as "there exists X such that X is Aristotle and X is a man," where "there exists" functions as a quantifier and X represents a variable. This difference separates it from propositional logic, which lacks quantifiers and relational expressions.

‍

Fluent

A state or attribute that varies with the passing of time. In logical methodologies for reasoning about actions, fluents can be expressed in first-order logic through predicates that include a time-dependent argument.

‍

Forward Chaining

One of the two primary reasoning techniques employed in an inference engine, forward chaining can be logically characterized as the repeated application of modus ponens. It is a widely used implementation approach in expert systems, business applications, and production rule frameworks. The inverse method of forward chaining is backward chaining. Forward chaining begins with known information and applies inference rules to derive additional facts, such as from an end user, until a target outcome is achieved. An inference engine utilizing forward chaining scans through inference rules to identify one where the premise is confirmed as true. Once such a rule is located, the system deduces or infers the conclusion, thereby incorporating new knowledge into its dataset.

‍

Foundational Model

A core model serving as the foundation for a solution set, generally pre-trained on vast datasets through self-supervised machine learning. Other models or applications are built upon foundational models or adapted into fine-tuned, context-specific variants. Examples include BERT, GPT-n, Llama, and DALL-E.

‍

Frame

A data structure in artificial intelligence designed to segment knowledge into smaller substructures by depicting typical scenarios. Frames serve as the fundamental data representation format in AI frame-based languages.

‍

Frame Language

A methodology for representing knowledge in AI. Frames are organized as ontologies comprising groups and subgroups of conceptual frames. They resemble class hierarchies in object-oriented programming languages, though their core objectives differ. Frames emphasize clear and intuitive knowledge representation, whereas objects prioritize encapsulation and data concealment. Frames emerged from AI research, while objects originated in software engineering. However, in practical applications, the functionalities and methodologies of frame-based and object-oriented languages substantially intersect.

‍

Frame Problem

The challenge of identifying suitable sets of axioms to provide a comprehensive and functional representation of a robotic environment.

‍

Friendly Artificial Intelligence

A theoretical artificial general intelligence designed to benefit humanity. It falls within the domain of AI ethics and is closely linked to machine ethics. While machine ethics addresses how an intelligent system ought to act, research on benevolent AI focuses on the practical implementation of such behavior and ensuring it remains appropriately regulated.

‍

F-Score (F-Measure, F1 Measure)

An F-score represents the harmonic mean of a system's precision and recall metrics. It is computed using the formula: 2 × [(Precision × Recall) / (Precision + Recall)]. A key critique of using F-score values to assess the effectiveness of a predictive system is that a relatively high F-score may result from an imbalance between precision and recall, failing to provide a complete picture. Additionally, highly accurate systems often face difficulty in enhancing precision or recall without adversely affecting the other.

‍

For critical applications where information retrieval is prioritized over accuracy, resulting in many false positives but ensuring that all true positives are identified, a different metric, known as the F2 measure, is used, placing greater emphasis on recall. Conversely, the F0.5 measure gives more weight to precision.

‍

Futures Studies

The exploration of theorizing potential, likely, and desirable futures, along with the perspectives and narratives that shape them.

‍

Fuzzy Control System

A regulatory mechanism utilizing fuzzy logic or a computational framework that interprets analog input magnitudes through logical variables that assume continuous values ranging from 0 to 1, as opposed to traditional or binary logic, which functions with discrete states of either 1 or 0, corresponding to true or false, respectively.

‍

Fuzzy Logic

A basic variant of many-valued logic, where the truth values of variables can assume any degree of truthfulness, represented by any real number within the inclusive range of 0, denoting Completely False, to 1, signifying Completely True. As a result, it is utilized to manage the notion of partial truth, wherein the truth value can span between entirely true and wholly false. Fuzzy logic contrasts with Boolean logic, in which variables can only take on integer values of 0 or 1.

‍

Fuzzy Rule

A guideline applied in fuzzy logic systems to deduce an outcome based on input parameters.

‍

Fuzzy Set

In traditional set theory, the inclusion of elements within a set is evaluated in a binary manner based on a two-valued condition. An element either belongs to the set or it does not. In contrast, fuzzy set theory allows for a gradual evaluation of element membership within a set, which is represented using a membership function that takes values within the real unit interval [0,1]. Fuzzy sets extend classical sets, as the characteristic functions of classical sets are specific instances of fuzzy set membership functions when restricted to values of 0 or 1. In fuzzy set theory, conventional two-valued sets are often referred to as crisp sets. This approach is applicable in various fields where data is incomplete or imprecise, such as bioinformatics.

Game Theory

The analysis of mathematical frameworks that model strategic interactions among rational decision-makers.

‍

General Game Playing (GGP)

General game playing involves creating AI systems capable of executing and competently engaging in multiple games.

‍

Generalization

The notion that humans, other animals, and artificial neural networks apply prior knowledge to current learning scenarios when the circumstances are perceived as analogous.

‍

Generalization Error

In supervised learning tasks within machine learning and statistical learning theory, generalization error, also referred to as out-of-sample error or risk, quantifies how precisely a learning algorithm can forecast results for data it has not encountered before.

‍

Generalized Model

A model that is not tailored to particular applications or specific types of information.

‍

Generative Adversarial Network (GAN)

A category of machine learning models where two neural networks compete against one another within a zero-sum game structure.

‍

Generative AI (GenAI)

Generative AI is a form of technology that leverages artificial intelligence to produce content such as text, video, code, and images. A generative AI system is trained on vast amounts of data, allowing it to identify patterns for generating new material.

‍

Generative Pretrained Transformer (GPT)

A substantial language model built upon the transformer framework that produces text. Initially, it undergoes pretraining to forecast the next token in sequences. A token is usually a word, subword, or punctuation mark. Following this pretraining phase, GPT models are capable of producing text resembling human writing by continually anticipating the most likely subsequent token. These models are typically further refined, for instance, through reinforcement learning guided by human input, to minimize inaccuracies or undesirable outputs, or to tailor the responses for conversational interactions.

‍

Generative Summarization

Leveraging large language model capabilities to process text-based inputs such as extended conversations, emails, reports, contracts, and policies, extracting key information and condensing it into essential summaries for rapid understanding. The process involves utilizing pretrained language models and contextual comprehension to generate brief, precise, and pertinent summaries.

‍

Genetic Algorithm (GA)

A metaheuristic based on the concept of natural selection falling under the broader category of evolutionary algorithms (EA). Genetic algorithms are frequently employed to produce optimal solutions for optimization and search issues, utilizing bio-inspired techniques such as mutation, crossover, and selection.

‍

Genetic Operator

An operator utilized in genetic algorithms to steer the algorithm toward a solution for a specific problem. There are three primary types of operators—mutation, crossover, and selection—which must function together in order for the algorithm to be effective.

‍

Glowworm Swarm Optimization

A swarm intelligence optimization algorithm inspired by the behavior of glowworms, also referred to as fireflies or lightning bugs.

‍

Gradient Boosting

A machine learning method based on boosting in a functional space, where the objective is pseudo-residuals rather than residuals as in conventional boosting.

‍

Graph (Abstract Data Type)

A graph is an abstract data structure designed to represent the concepts of undirected and directed graphs from mathematics, particularly within the domain of graph theory.

‍

Graph (Discrete Mathematics)

In mathematics, particularly in graph theory, a graph is a structure consisting of a collection of objects where certain pairs of these objects are connected in some way. The objects are referred to as vertices, also known as nodes or points, and each connected pair of vertices is called an edge, also known as an arc or line.

‍

Graph Database (GDB)

A database that employs graph structures for semantic queries, using nodes, edges, and attributes to represent and store information. The central concept of the system is the graph, or edge or relationship, which directly connects data elements in the store, a collection of data nodes and edges representing the connections between those nodes. These relationships support the direct linking of data within the store, often allowing retrieval with a single operation. Graph databases prioritize the relationships between data, making querying these relationships efficient, as they are persistently stored within the database. Relationships can be easily visualized using graph databases, making them especially useful for highly interconnected data.

‍

Graph Theory

The analysis of graphs, which are mathematical constructs used to represent pairwise connections between entities.

‍

Graph Traversal

The procedure of visiting, examining, and/or modifying each node in a graph. These traversals are categorized by the sequence in which the nodes are visited. Tree traversal is a specific instance of graph traversal.

‍

Grounding

The capability of generative systems to trace the factual content within a generated output or completion. It connects generative applications to accessible factual sources, such as documents or knowledge repositories, either by providing a direct citation or by searching for additional references.

‍

Guardrails

Guardrails refer to constraints and guidelines imposed on AI systems to ensure they process data responsibly and do not produce unethical content.

Hallucination

Hallucination describes an inaccurate response from an AI system or misleading information in an output that is presented as if it were true.

‍

Hallucitations

Fictitious data that consists of invented, incorrect, or misleading references or sources mistakenly presented as factual within generated content.

‍

Heuristic

A method developed to solve a problem more rapidly when traditional approaches are too sluggish, or to find an approximate solution when traditional methods cannot identify an exact one. It is accomplished by sacrificing optimality, completeness, accuracy, or precision in favor of speed. In essence, it can be seen as a shortcut. A heuristic function, often referred to simply as a heuristic, is a function that ranks options in search algorithms at each decision point based on the available data to determine which path to pursue. For instance, it may estimate the precise solution.

‍

Hidden Layer

A layer of neurons in an artificial neural network that is neither an input layer nor an output layer.

‍

Human-Centered Perspective

A human-focused viewpoint envisions AI systems collaborating with individuals and enhancing human abilities. Humans should always maintain a primary role in education, and AI should not serve as a substitute for teachers.

‍

Hybrid AI

Hybrid AI refers to any artificial intelligence system that integrates multiple AI approaches. In natural language processing, this typically involves utilizing both symbolic reasoning and machine learning techniques within a single workflow.

‍

Hyper-Heuristic

A heuristic search approach that aims to automate the process of choosing, merging, creating, or modifying multiple simpler heuristics, or elements of such heuristics, to effectively resolve computational search issues, often through the integration of machine learning methods. One of the driving factors behind exploring hyper-heuristics is to develop systems capable of addressing categories of problems instead of focusing on solving just a single problem.

‍

Hyperparameter

A hyperparameter is a variable or setting that influences how an AI model learns. It is typically configured manually outside the model.

‍

Hyperparameter Optimization

The procedure of selecting an ideal set of hyperparameters for a learning algorithm.

‍

Hyperplane

A decision boundary in machine learning classifiers that divides the input space into multiple regions, with each region representing a distinct class label.

IEEE Computational Intelligence Society

A professional organization within the Institute of Electrical and Electronics Engineers (IEEE) concentrating on the principles, design, implementation, and advancement of biologically and linguistically inspired computational frameworks, emphasizing neural networks, connectionist systems, genetic algorithms, evolutionary programming, fuzzy systems, and hybrid intelligent systems incorporating these frameworks.

‍

Image Recognition

Image recognition is the process of detecting and identifying objects, individuals, locations, or text within an image or video.

‍

Incremental Learning

A machine learning approach in which input data is progressively utilized to expand the current model's understanding, i.e., to continue training the model. It signifies a flexible method of supervised and unsupervised learning that can be employed when training data is provided incrementally or when its volume exceeds system memory capacity. Algorithms capable of supporting this incremental learning process are referred to as incremental machine learning algorithms.

‍

Inference Engine

A module within an expert system that utilizes logical principles to analyze the knowledge base and infer new or supplementary information.

‍

Information Integration (II)

The integration of data from several sources with varying conceptual, contextual, and typographical formats. It is employed in data mining and the consolidation of information from unstructured or semi-structured sources. While information integration often pertains to textual forms of knowledge, it can also be applied to multimedia content. A related concept, information fusion, refers to the amalgamation of data into a new set of information aimed at minimizing redundancy and uncertainty.

‍

Information Processing Language (IPL)

A programming language designed with features to assist in creating programs that carry out basic problem-solving tasks, such as lists, dynamic memory allocation, data types, recursion, functions as parameters, generators, and cooperative multitasking. IPL introduced the concept of list processing, although in an assembly-language format.

‍

Insight Engine

An insight engine, also known as cognitive search or enterprise knowledge discovery, employs relevance-based techniques to interpret, uncover, structure, and examine data. It merges search functionality with AI to supply information for users and datasets for machines. The primary objective of an insight engine is to deliver timely data that yields actionable insights.

‍

Intelligence Amplification (IA)

The efficient application of information technology in enhancing human intellect.

‍

Intelligence Explosion

A potential consequence of humanity creating artificial general intelligence. AGI would possess the ability to recursively improve itself, resulting in the swift rise of artificial superintelligence, the extent of which is uncertain, at the moment of the technological singularity.

‍

Intelligent Agent (IA)

A self-sufficient entity that performs actions, focusing its efforts on reaching objectives, which means it is an agent. The entity functions within an environment by perceiving through sensors and responding with actuators indicating intelligence. Intelligent agents can also acquire knowledge or use existing information to fulfill their objectives. They can range from very basic to highly intricate.

‍

Intelligent Control

A category of control methods that employ different artificial intelligence computing strategies such as neural networks, Bayesian probability, fuzzy logic, machine learning, reinforcement learning, evolutionary computation, and genetic algorithms.

‍

Intelligent Document Processing (IDP)/Intelligent Document Extraction and Processing (IDEP)

The capability to automatically interpret and transform unstructured and semi-structured data, recognize relevant information, extract it, and utilize it through automated workflows. Intelligent Document Processing is frequently a foundational technology for Robotic Process Automation (RPA) operations.

‍

Intelligent Personal Assistant

A software agent capable of carrying out tasks or providing services for a user based on spoken instructions. The term chatbot is sometimes used to describe virtual assistants broadly or specifically those accessed through online chat; or, in some cases, chat programs designed solely for entertainment. Some virtual assistants have the ability to understand spoken language and reply with synthesized voices. Users can ask their assistants questions, control smart home devices and media playback using text or voice commands, and manage simple tasks like email, to-do lists, and calendars with verbal instructions.

‍

Intelligent Tutoring Systems (ITS)

A computational platform or digital educational environment that provides immediate and personalized feedback to learners. An Intelligent Tutoring System can utilize rule-based AI, where rules are defined by a human or employ machine learning behind the scenes. By behind the scenes, we refer to the foundational algorithms and code that form the ITS. These systems can facilitate adaptive learning.

‍

Interpretable Machine Learning (IML)

Interpretable machine learning, also known as interpretable AI, refers to the development of models that are naturally understandable, as they offer built-in explanations for their decisions. This method is favored over explainable machine learning for various reasons, including the principle that we should comprehend how our systems function from the outset rather than attempting to clarify opaque models afterward.

‍

Interpretation

An attribution of meaning to the symbols of a formal language. Numerous formal languages employed in mathematics, logic, and theoretical computer science are defined purely in syntactic terms, meaning they hold no meaning until a specific interpretation is provided. The broader study of the interpretations of formal languages is referred to as formal semantics.

‍

Intrinsic Motivation

An intelligent agent is internally driven to act as if the mere informational value of the outcome of the action serves as the motivating element. In this context, information content is quantified in terms of information theory, measuring uncertainty. A common form of intrinsic motivation is the pursuit of novel or unexpected scenarios, unlike typical extrinsic motivations such as seeking food. Artificial agents guided by intrinsic motivation exhibit behaviors similar to exploration and curiosity.

‍

Issue Tree

A visual representation of a question that breaks it down into its various elements vertically, with increasing detail as it moves to the right. Issue trees are helpful in problem-solving to pinpoint the underlying causes of an issue and to discover possible solutions. They also serve as a guide to understanding how each part contributes to the overall understanding of the problem.

Junction Tree Algorithm

A technique employed in machine learning to derive marginalization in general graphs. Essentially, it involves executing belief propagation on a transformed graph known as a junction tree. The graph is termed a tree due to its branching structure, where the nodes of variables represent the branches.

Kernel Method

In machine learning, kernel techniques are a category of algorithms used for pattern recognition, with the most well-known being the support vector machine, commonly abbreviated as SVM. The primary goal of pattern recognition is to identify and examine general types of relationships within datasets. Examples of these general types include cluster analysis, rankings, principal components, correlations, and classifications.

‍

KL-ONE

A renowned knowledge representation framework in the tradition of semantic networks and frames; in other words, it is a frame-based language. The system seeks to address semantic ambiguity in semantic network models and to clearly represent conceptual knowledge as an organized inheritance structure.

‍

K-Means Clustering

A technique of vector quantization, initially derived from signal processing, that seeks to divide n data points into k groups, with each data point assigned to one of the groups. The closest average, known as a cluster center or centroid, acts as the representative of the cluster.

‍

K-Nearest Neighbors

A non-parametric supervised learning technique initially created by Evelyn Fix and Joseph Hodges in 1951 and later improved by Thomas Cover. It is utilized for both classification and regression tasks.

‍

Knowledge Acquisition

The procedure used to establish the rules and ontologies needed for a knowledge-based system. The term was originally employed in relation to expert systems to describe the initial steps involved in creating an expert system, such as identifying and consulting domain specialists and recording their expertise through rules, objects, and frame-based ontologies.

‍

Knowledge-Based AI

Knowledge-based systems (KBS) are a type of AI built to encapsulate the expertise of human specialists, aiding in decision-making and resolving problems.

‍

Knowledge-Based System

A software application that uses reasoning and a knowledge repository to address intricate issues. The term is expansive and encompasses various types of systems. The shared principle between all knowledge-based systems is the effort to represent knowledge clearly and a reasoning mechanism that supports the generation of new knowledge. Consequently, a knowledge-based system has two key characteristics: a knowledge repository and a reasoning engine.

‍

Knowledge Distillation

The method of transferring knowledge from a larger machine learning model to a more compact one.

‍

Knowledge Engineering (KE)

A technique for helping computers emulate human-like understanding. Knowledge engineers embed reasoning into knowledge-based systems by gathering, structuring, and incorporating general or specialized expertise into a framework.

‍

Knowledge Extraction

The process of generating knowledge from structured relational databases, XML, and unstructured sources such as text, documents, and images. The resulting knowledge must be in a machine-readable and interpretable format and should be structured to support reasoning. While similar to information extraction and ETL, the key distinction is that the output must extend beyond merely creating structured data or transforming it into a relational schema. It involves either repurposing existing formal knowledge, such as creating a schema based on the source data or reusing identifiers or ontologies.

‍

Knowledge Graph

A knowledge graph is a network of interconnected concepts whose significance lies in its capacity to accurately depict a segment of reality, whether general or specialized. Each concept is linked to at least one other, with the nature of these connections categorized into different types.

The meaning of each concept is defined by its relationships, making every node representative of its concept solely based on its placement within the graph. Advanced knowledge graphs may associate numerous attributes with a node, including the terms used in language to describe the concept, its cultural connotations, and its grammatical behavior in a sentence.

‍

Knowledge Interchange Format (KIF)

A programming language created to allow systems to exchange and reuse information from knowledge-based systems. KIF is comparable to frame languages like KL-ONE and LOOM, but unlike these languages, its main purpose is not to serve as a framework for expressing or utilizing knowledge. Instead, it is designed for the exchange of knowledge between systems. The creators of KIF compared it to PostScript, which was not primarily developed as a tool for storing and manipulating documents but as a format for systems and devices to share documents. Similarly, KIF is intended to facilitate the sharing of knowledge between separate systems that may use different languages, formalisms, platforms, and so on.

‍

Knowledge Model

A method for developing a machine-readable representation of knowledge or guidelines related to a language, domain, or set of processes. This representation is structured in a data format that allows the information to be stored in a database and understood by software.

‍

Knowledge Representation and Reasoning (KR² or KR&R)

The branch of artificial intelligence focused on representing information about the world in a format that a computer system can use to tackle intricate tasks, such as diagnosing medical conditions or engaging in conversations in a natural language. Knowledge representation draws on psychological research about how humans approach problem-solving and store knowledge, aiming to create frameworks that make it easier to design and develop complex systems. It also incorporates insights from logic to automate different types of reasoning, such as applying rules or analyzing the relationships between sets and subsets. Examples of knowledge representation frameworks include semantic networks, system architectures, frames, rules, and ontologies. Automated reasoning tools include inference engines, theorem provers, and classifiers.

LangOps or Language Operations

The procedures and methodologies that facilitate the development, formulation, evaluation, implementation, and continuous refinement of linguistic models and natural language solutions.

‍

Language Data

Linguistic data consists of words and constitutes a type of unstructured information. This qualitative data, also called textual data, essentially pertains to the written and spoken expressions in a language.

‍

Language Model

A stochastic model that processes natural language.

‍

Large Language Model (LLM)

A large language model is an AI system trained on vast quantities of text, allowing it to comprehend language and produce human-like text.

‍

Lazy Learning

In machine learning, lazy learning is a technique where the generalization of the training data is, theoretically, postponed until a query is presented to the system, unlike eager learning, where the system attempts to generalize the training data prior to receiving any queries.

‍

Lemma

The root form of a word that serves as the representation for all its conjugated or declined variations.

‍

Lexicon

Awareness of all potential interpretations of words within their appropriate context is essential for accurately analyzing textual content with a high degree of precision.

‍

Limited Memory

Limited memory is a category of AI systems that gathers information from real-time events and retains it in a database to improve predictions.

‍

Linked Data

Interconnected data is a term that indicates whether a recognizable repository of information is associated with another. It is commonly utilized as a standardized reference. For example, a knowledge graph where each concept or node is connected to its corresponding entry on Wikipedia.

‍

Lisp Programming Language (LISP)

A group of programming languages with an extensive history and a unique, completely parenthesized prefix syntax.

‍

Logic Programming

A kind of programming paradigm primarily grounded in formal logic. Any program created in a logic programming language consists of a collection of statements in logical form, representing facts and rules about a specific problem domain. Prominent families of logic programming languages include Prolog, answer set programming (ASP), and Datalog.

‍

Long Short-Term Memory (LSTM)

A type of artificial recurrent neural network architecture employed in deep learning. In contrast to typical feedforward neural networks, LSTM incorporates feedback connections to function as a universal computer. This feature allows the system to perform any computation that a Turing machine can execute. It is capable of processing not just individual data points, like images, but also complete sequences of data, such as speech or video.

Machine Learning (ML)

Machine learning is a branch of AI that integrates elements of computer science, mathematics, and programming. It emphasizes creating algorithms and models that help machines learn from data and forecast patterns and behaviors without human intervention.

‍

Machine Listening

A broad area of research focused on algorithms and systems for machine-based audio comprehension.

‍

Machine Perception

The ability of a computer system to analyze data in a way that mirrors how humans use their senses to interact with and understand the environment around them.

‍

Machine Vision (MV)

The ability of a computer system to analyze data in a manner similar to how humans use their senses to understand and interact with the world. This technology and its methods are employed to provide imaging-based automatic inspection and analysis in applications such as automated examination, process management, and robotic navigation, primarily in industrial settings. Machine vision is an umbrella term that covers a wide range of technologies, software, hardware products, integrated systems, techniques, and expertise. As a systems engineering field, machine vision is often regarded as separate from computer vision. Machine vision seeks to integrate existing technologies in innovative ways to tackle real-world challenges. While the term is most commonly used in industrial automation, it is also applied in other fields like security and vehicle navigation.

‍

Markov Chain

A probabilistic model that represents a series of potential events, where the likelihood of each event is determined solely by the condition reached in the immediately preceding event.

‍

Markov Decision Process (MDP)

A discrete-time probabilistic control process. It offers a mathematical structure for representing decision-making in scenarios where results are partially random and partially influenced by the choices of a decision-maker. MDPs are valuable for analyzing optimization problems addressed through dynamic programming and reinforcement learning.

‍

Mathematical Optimization

A discrete-time stochastic control process. It provides a mathematical framework for modeling decision-making in situations where outcomes are partly random and partly shaped by the decisions of a chooser. MDPs are useful for examining optimization challenges tackled through dynamic programming and reinforcement learning.

‍

Mechanism Design

A domain within economics and game theory that adopts an engineering perspective to create economic mechanisms or incentives aimed at achieving specific goals in strategic environments where participants behave rationally. Since it begins with the end result and works backward, it is also referred to as reverse game theory. It has wide-ranging uses in economics, politics, markets, auctions, voting methods, and networked systems. Networked systems include internet interdomain routing and sponsored search auctions.

‍

Mechatronics

A cross-disciplinary field of engineering that concentrates on the design and development of both electrical and mechanical systems. It incorporates elements of robotics, electronics, computing, telecommunications, systems, control, and product engineering.

‍

Metabolic Network Reconstruction and Simulation

Provides a comprehensive understanding of the molecular processes within a specific organism. Specifically, these models link the genome to molecular physiology.

‍

Metacontext and Metaprompt

Core guidelines on structuring the training process to shape the model's expected behavior.

‍

Metadata

Information that characterizes or offers details regarding other data.

‍

Metaheuristic

A metaheuristic is an advanced method or strategy intended to discover, create, or choose a heuristic. The partial search algorithm of the heuristic aids in solving problems with insufficient or imperfect data or constrained computational resources. Metaheuristics explore a set of solutions that is too extensive to be fully sampled.

‍

Model

A machine learning model is the resulting construct generated once an ML algorithm has processed the provided sample data during its training stage. This model is subsequently utilized by the algorithm in deployment to interpret text in NLP scenarios and deliver insights and/or forecasts.

‍

Model Checking

Model checking or property verification involves thoroughly and automatically assessing whether a specific model of a system adheres to a specified criterion. Usually, this pertains to hardware or software systems, with the specification including safety conditions like the prevention of deadlocks and other critical situations that might lead to system failure. Model checking is a method for autonomously validating the correctness attributes of finite-state systems.

‍

Model Drift

Model drift refers to the deterioration of a model's predictive accuracy due to shifts in real-world conditions. This decline occurs for various reasons, such as alterations in the digital landscape and subsequent changes in the relationships between variables. For instance, a spam detection model trained on email content may lose effectiveness if the patterns in spam emails evolve.

‍

Model Parameter

These are adjustable factors within the model that are established through training data. They represent the configured or optimized internal variables whose values can be inferred from data. These elements are essential for the model when generating predictions, as their values determine the model's effectiveness and alignment with the data.

‍

Modus Ponens

In propositional logic, modus ponens is an inference rule. It can be expressed as "P implies Q, and P is affirmed to be true, so Q must also be true."

‍

Modus Tollens

In propositional logic, modus tollens is a legitimate form of argument and an inference rule. It applies the general principle that if a statement is true, then its contrapositive must also be true. The inference rule modus tollens asserts that the reasoning from P implies Q to the negation of Q, implying the negation of P is valid.

‍

Monte Carlo Tree Search

Monte Carlo tree search (MCTS) is a heuristic search technique used for certain types of decision-making processes.

‍

Morphological Analysis

Decomposing a problem that has numerous established solutions into its fundamental components or simplest forms to achieve a deeper understanding. Morphological analysis is applied in broad problem-solving, language studies, and biological sciences.

‍

Multi-Agent System (MAS)

A computerized framework consisting of several interacting intelligent agents. Multi-agent systems can tackle challenges that are too complex or unachievable for a single agent or a unified system. Intelligence may encompass systematic, functional, procedural methods, algorithmic exploration, or reinforcement learning.

‍

Multilayer Perceptron (MLP)

In deep learning, a multilayer perceptron refers to a contemporary feedforward neural network made up of fully connected neurons with nonlinear activation functions arranged in layers, recognized for its ability to differentiate data that cannot be separated linearly.

‍

Multimodal Models and Modalities

Linguistic models trained to comprehend and process various data modalities, including text, visuals, sound, and other formats, leading to better performance within a broader spectrum of tasks.

‍

Multi-Swarm Optimization

A variation of particle swarm optimization (PSO) that utilizes multiple sub-swarms rather than a single standard swarm. The general strategy in multi-swarm optimization is for each sub-swarm to concentrate on a particular area, while a specific diversification technique determines when and where to deploy the sub-swarms. The multi-swarm approach is particularly suited for optimization in multimodal issues, where several local optima are present.

‍

Multitask Prompt Tuning (MPT)

A method that structures a prompt as a modifiable variable to enable repeated prompts where only the variable is adjusted.

‍

Mutation

A genetic operator employed to preserve genetic diversity between generations of a population in a genetic algorithm. It is similar to biological mutation. Mutation modifies one or more gene values in a chromosome from its original state. With mutation, the solution may completely differ from the previous one. As a result, the genetic algorithm can potentially find a better solution by using mutation. Mutation takes place during evolution based on a user-defined mutation probability, which should be kept low. If the probability is too high, the search may devolve into a basic random search.

‍

MYCIN

An early backward chaining expert system that utilized artificial intelligence to identify bacteria responsible for serious infections, such as bacteremia and meningitis, and to suggest antibiotics, with dosage adjusted according to the patient's body weight. The name originated from the antibiotics, as many of them end with the suffix -mycin. The MYCIN system was also employed to diagnose blood clotting disorders.

Naive Bayes Classifier

In machine learning, naive Bayes classifiers are a group of straightforward probabilistic classifiers that rely on applying Bayes' theorem with strong assumptions of independence among the features.

‍

Naive Semantics

A method utilized in computer science for encoding fundamental knowledge about a particular domain. It has been applied toward understanding the meaning of natural language statements in AI systems. In a broader context, the term has been used to describe the application of a restricted set of commonly recognized knowledge about a specific area of the world. It has been employed in fields like the knowledge-based design of data schemas.

‍

Name Binding

In programming languages, name binding refers to the linking of entities, which can be either data or code, with identifiers. An identifier linked to an object is said to reference that object. Machine languages do not inherently include the concept of identifiers, but programming languages implement name-object bindings as a service and notation for the programmer. Binding is closely related to scoping, as the scope defines which names are associated with which objects, determining the locations in the program code and the execution paths in which these associations occur. The use of an identifier id in a context that establishes a binding for id is called a binding occurrence. In all other instances, an identifier represents what it is associated with; these are called applied occurrences. Examples of occurrences include expressions, assignments, and subroutine calls.

‍

Named-Entity Recognition (NER)

A subset of information extraction focused on identifying and categorizing named entity references in unstructured text into predetermined categories, such as personal names, organizations, places, medical codes, temporal expressions, quantities, financial values, percentages, and more.

‍

Named Graph

A fundamental idea of Semantic Web architecture, where a collection of Resource Description Framework (RDF) statements is identified using a URI, enabling the description of that collection with additional information such as context, provenance, or other types of metadata. Named graphs are a basic expansion of the RDF data model that allows the creation of graphs, though the model lacks a practical method for differentiating between them once they are made publicly available on the Web.

‍

Natural Language Generation (NLG)

Systems that autonomously transform organized data, such as that stored in a database, an application, or a real-time stream, into a text-based narrative. This ability expands user accessibility by allowing information to be read or heard, thereby improving understanding.

‍

Natural Language Processing (NLP)

Natural language processing is a branch of AI that allows computers to interpret and comprehend both spoken and written human language. NLP powers functionalities such as speech and text recognition on various devices. 

‍

Natural Language Programming

A method of programming supported by ontology, expressed through natural-language sentences, such as English.

‍

Natural Language Query (NLQ)

A natural language input consisting solely of words and expressions as they appear in verbal communication, excluding any non-linguistic symbols or characters.

‍

Natural Language Technology (NLT)

A specialized branch of linguistics, computer science, and AI that focuses on natural language processing, natural language understanding, and natural language generation.

‍

Natural Language Understanding (NLU)

A branch of natural language processing, natural language understanding concentrates on the genuine machine interpretation of examined and structured unstructured linguistic data. The process is facilitated through semantics.

‍

Network Motif

All types of networks, including biological, social, technological, and others, can be depicted as graphs, which encompass a broad range of subgraphs. Examples of technological networks involve computer systems and electrical circuits. A key local characteristic of networks is the concept of network motifs, which are identified as recurring and statistically meaningful subgraphs or patterns.

‍

Neural Machine Translation (NMT)

A method of machine translation that employs a vast ANN to estimate the probability of a word sequence, usually representing complete sentences within a unified model.

‍

Neural Network

A neural network is a deep learning method modeled after the structure of the human brain. These networks rely on vast data sets to carry out computations and generate outputs, supporting capabilities such as speech and image recognition.

‍

Neural Turing Machine (NTM)

A recurrent neural network architecture. NTMs merge the flexible pattern recognition abilities of neural networks with the computational strength of programmable systems. An NTM features a neural network controller linked to external memory resources, which it engages with via attention mechanisms. These memory interactions are differentiable throughout, allowing for optimization through gradient descent. An NTM with a long short-term memory (LSTM) controller can deduce basic algorithms like copying, sorting, and associative recall solely from examples.

‍

Neurocybernetics

A direct communication channel between an augmented or wired brain and an external device. BCI distinguishes itself from neuromodulation by supporting two-way information exchange. BCIs are frequently focused on studying, mapping, supporting, enhancing, or restoring human cognitive or sensory-motor abilities.

‍

Neuro-Fuzzy

Mergers of artificial neural networks and fuzzy reasoning.

‍

Neuromorphic Engineering

A concept referring to the application of very-large-scale integration (VLSI) systems incorporating electronic analog circuits to replicate neuro-biological structures found in the nervous system. Recently, the term neuromorphic has been applied to describe analog, digital, hybrid analog/digital VLSI, and software systems that model neural systems. These systems manage tasks such as perception, motor regulation, or multisensory coordination. The realization of neuromorphic computing at the hardware level can be achieved using oxide-based memristors, spintronic memories, threshold switches, and transistors.

‍

Node

A fundamental component of a data structure, such as a linked list or tree structure. Nodes store data and may also connect to other nodes. Connections between nodes are commonly represented by pointers.

‍

Nondeterministic Algorithm

A procedure that, even with identical input, can display varying behaviors on different executions, in contrast to a deterministic algorithm.

‍

Nouvelle AI

Nouvelle AI distinguishes itself from traditional AI by seeking to create robots with intelligence comparable to that of insects. Researchers contend that intelligence can arise naturally from basic behaviors as these intelligences interact with the real world rather than relying on the artificial environments that symbolic AIs usually require to be programmed into them.

‍

NP

In computational complexity theory, NP, or nondeterministic polynomial time, is a classification category used to group decision problems. NP consists of decision problems for which instances where the answer is "yes" have solutions that can be verified in polynomial time.

‍

NP-Completeness

In computational complexity theory, a problem is considered NP-complete when it can be solved by a limited class of exhaustive search algorithms and can be used to replicate any other problem with a comparable algorithm. More specifically, each input to the problem must be linked to a set of solutions of polynomial size, whose correctness can be verified efficiently in polynomial time, ensuring that the result for any input is "yes" if the solution set is non-empty and "no" if it is empty.

‍

NP-Hardness

Also known as nondeterministic polynomial-time difficulty, in computational complexity theory, it refers to a characteristic of a group of problems that are, informally, at least as challenging as the most difficult problems in NP. A basic illustration of an NP-hard problem is the subset sum problem.

Occam's Razor

The problem-solving guideline asserting that when faced with rival hypotheses yielding identical predictions, one should opt for the explanation requiring the least assumptions. This principle is not intended to eliminate hypotheses that produce distinct forecasts. The concept is credited to the English Franciscan friar William of Ockham (circa 1287–1347), a scholastic thinker and theologian.

‍

Offline Learning

A machine learning instructional method where a model is trained using a static dataset that remains unchanged throughout the training procedure.

‍

Online Machine Learning

A machine learning approach where data arrives in a sequential manner and is utilized to incrementally refine the optimal predictor for future instances at each step, in contrast to batch learning methods that derive the best predictor by processing the entire training dataset simultaneously. Online learning is frequently employed in domains of machine learning where training on the full dataset is computationally impractical, necessitating the use of out-of-core algorithms. Additionally, it is applied in scenarios requiring the algorithm to continuously adjust to emerging patterns in the data or when the data itself evolves over time.

‍

Ontology

An ontology resembles a taxonomy but expands its basic hierarchical classification framework by incorporating attributes for each node/element and establishing links between nodes that may extend along different branches. These attributes are neither uniform nor confined to a fixed set, requiring mutual agreement between the classifier and the user.

‍

Ontology Learning

The fully or partially automated construction of ontologies, encompassing the extraction of relevant domain-specific terminology and the associations between the concepts these terms signify from a collection of natural language text, followed by their encoding using an ontology language to facilitate efficient retrieval.

‍

OpenAI

The profit-driven entity OpenAI LP, overseen by its non-profit parent organization OpenAI Inc., carries out investigations in the domain of artificial intelligence with the declared objective of fostering and advancing benevolent AI in a manner that serves the collective welfare of humankind.

‍

OpenCog

A venture focused on developing an open-source framework for AI. OpenCog Prime is a structural design for robotic and virtual embodied cognition, outlining a collection of interdependent modules intended to foster human-level artificial general intelligence as an emergent property of the entire system.

‍

Open Mind Common Sense

An AI initiative housed at the Massachusetts Institute of Technology (MIT) Media Lab aimed at constructing and leveraging an extensive repository of commonsense knowledge and compiled through the collective input of numerous individuals on the internet.

‍

Open-Source Software (OSS)

A category of computer programs where the source code is made available under a license that permits the copyright owner to authorize users to examine, modify, and share the software with anyone for any purpose. Open-source software can be created through a cooperative and publicly accessible process. It serves as a notable instance of open collaboration.

‍

Overfitting

Overfitting happens during machine learning training when the algorithm becomes too specialized in the training data and struggles to handle new examples. A well-functioning AI model should be able to recognize patterns in the data and apply them to unfamiliar tasks.

Parameters

A collection of numerical coefficients representing neural linkages or other components within an AI model, with values established through training. Extensive language models can contain billions of such parameters.

‍

Parsing

Recognizing the individual components that make up a text and then attributing to them their corresponding logical and grammatical significance.

‍

Partially Observable Markov Decision Process (POMDP)

A broader version of a Markov decision process (MDP). A POMDP represents an agent's decision-making process where it is assumed that the system dynamics are governed by an MDP, but the agent cannot directly perceive the underlying state. Instead, the agent must keep a probability distribution over the possible states, relying on a series of observations and observation probabilities, as well as the underlying MDP.

‍

Partial Order Reduction

A method for minimizing the size of the state-space to be explored by a model checking or automated planning and scheduling algorithm. It takes advantage of the commutativity of simultaneously performed transitions, which produce the same state regardless of the order in which they are executed.

‍

Particle Swarm Optimization (PSO)

A computational technique that perfects a problem by repetitively attempting to refine a potential solution based on a specific quality measure. It addresses the problem by using a group of potential solutions, referred to as particles, and adjusting their positions within the search-space based on straightforward mathematical formulas involving the particle's position and velocity. Each particle's movement is affected by its own best-known position but is also directed toward the best-known positions in the search-space, which are updated as superior positions are discovered by other particles. This process is anticipated to guide the swarm toward the optimal solutions.

‍

Part-of-Speech Tagging

A Part-of-Speech (POS) tagger is an NLP feature that determines the grammatical attributes of words within a sentence. Basic POS tagging may simply classify each word by its grammatical category, whereas more advanced versions can cluster phrases and sentence components, distinguish various clause types, construct a dependency structure for a sentence, and even designate a logical role to each word.

‍

Pathfinding

Also known as pathing, the calculation, by a computer program, of the quickest route between two locations. It is a more applicable version of solving mazes. This area of study is largely based on Dijkstra's algorithm for determining the shortest path in a weighted graph.

‍

Pattern Recognition

Pattern recognition is the process of utilizing computational algorithms to examine, identify, and categorize recurring structures in data. This process helps determine how the data is sorted into various groups.

‍

Perceptron

A method for supervised training of binary classifiers.

‍

Plugins

A software module or extension that augments the capabilities of an LLM system in various settings, such as booking travel, online shopping, internet navigation, and performing mathematical computations.

‍

Post-Edit Machine Translation (PEMT)

The solution allows a translator to refine a document that has been previously translated by a machine. This process is usually performed on a sentence-by-sentence basis using a dedicated computer-aided translation tool.

‍

Post-Processing

Methods that may encompass different trimming techniques, rule refinement, or even knowledge incorporation. These processes act as a symbolic sieve for ambiguous and imprecise information generated by an algorithm.

‍

Precision

For a given set of results from a processed document, precision represents the percentage that denotes how many of those results are accurate according to the expectations of a particular application. This metric is relevant for many types of predictive AI systems, including search, classification, and entity detection.

For instance, imagine an application designed to identify all dog breeds mentioned in a document. If the document references 10 different dog breeds but the application retrieves only five, and all of them are correct, the system has achieved 100% precision. Even though some instances were overlooked, every identified result was accurate.

‍

Predicate Logic

A group of formal systems employed in philosophy, mathematics, linguistics, and computer science. First-order logic utilizes quantified variables over non-logical entities and permits the construction of sentences containing variables, such that instead of statements like "Aristotle is a man," one can express ideas in the form "there exists x such that x is Aristotle and x is a man," where "there exists" is a quantifier and x is a variable. This aspect differentiates it from propositional logic, which does not incorporate quantifiers or relationships; in this way, propositional logic serves as the basis for first-order logic.

‍

Predictive Analytics

Predictive analytics is a form of data analysis that leverages technology to forecast future events within a given timeframe by examining past data and trends.

‍

Pre-Processing

A phase in the data mining and analytical workflow that converts unprocessed information into a structured format that computers can interpret and evaluate. Processing structured data, such as numerical values, timestamps, monetary figures, and percentages, is relatively simple. However, unstructured data, including text and images, must first undergo cleansing and formatting before it can be effectively analyzed.

‍

Prescriptive Analytics 

Prescriptive analytics is a branch of data analysis that utilizes technology to evaluate factors such as potential scenarios, historical data, current performance, and available resources to assist organizations in making more informed strategic decisions.

‍

Pretrained Model

A model trained to perform a specific function, usually one that is applicable in many organizations or scenarios. Additionally, a pretrained model can serve as a foundation for developing a refined, context-adapted version, thereby supporting transfer learning.

‍

Pretraining

The initial phase in training a foundational model, typically conducted as an unsupervised learning stage. After pretraining, foundational models possess broad functionality. However, they require further refinement through fine-tuning to achieve higher precision.

‍

Principal Component Analysis (PCA)

A statistical method that applies an orthogonal transformation to convert a collection of observations of potentially correlated variables, or entities that each assume different numerical values, into a set of uncorrelated variables known as principal components. This transformation is structured so that the first principal component captures the maximum possible variance or explains the greatest amount of variability in the data, and each subsequent component, in turn, captures the highest variance possible while remaining orthogonal to the preceding components. The resulting vectors, each being a linear combination of the variables and containing n observations, form an uncorrelated orthogonal basis set. PCA is sensitive to the relative scaling of the original variables.

‍

Principle of Rationality

A concept introduced by Karl R. Popper in his 1963 Harvard Lecture, later published in his book Myth of Framework. It is connected to what he referred to as the "logic of the situation" in an article for Economica in 1944/1945, which was later included in his book The Poverty of Historicism. Popper's principle of rationality asserts that individuals behave in the most appropriate manner based on objective circumstances. This perspective represents an idealized view of human conduct that Popper employed to develop his model of situational logic.

‍

Probabilistic Programming (PP)

A coding framework where stochastic models are defined. Reasoning for these models is executed automatically. It seeks to integrate probabilistic modeling with conventional multipurpose programming to expand accessibility and broaden its applications. This approach can be employed to develop systems that assist in decision-making under uncertain conditions. Coding languages designed for probabilistic programming are known as Probabilistic Programming Languages (PPLs).

‍

Production System

A software application generally employed to deliver artificial intelligence, primarily composed of a collection of guidelines governing behavior, along with the framework required to adhere to those guidelines as the system reacts to various conditions in its environment.

‍

Programming Language

A structured linguistic system consisting of a collection of commands that generate different types of results. Coding languages are utilized in software development to execute computational procedures.

‍

Prolog

A rule-based coding language linked to machine intelligence and language processing. Prolog originates from predicate logic, a structured reasoning system, and, unlike numerous other coding languages, it is mainly designed as a declarative programming paradigm. The logic of the program is defined through relationships, represented as assertions and directives. Execution begins by posing a question against these relationships.

‍

Prompt

A prompt is a user-provided input given to an AI system to generate a specific response or output.

‍

Prompt Chaining

A method that employs a series of prompts to iteratively refine a query directed at a model.

‍

Prompt Engineering

The skill of structuring and refining user queries for an LLM or LLM-powered chatbot to obtain the most optimal outcome, typically accomplished through extensive trial and error. 

‍

Propositional Calculus

A field of reasoning that focuses on statements, which may be either true or false, and the progression of arguments. Complex statements are constructed by linking simpler ones using logical operators. Statements that lack logical operators are referred to as fundamental statements. Unlike predicate logic, statement logic does not involve non-logical entities, attributes describing them, or quantifiers. However, all the mechanisms of statement logic are encompassed within predicate logic and more advanced logical systems. In this regard, statement logic serves as the basis for predicate logic and higher-order logical frameworks.

‍

Proximal Policy Optimization (PPO)

An ML technique for optimizing an autonomous agent's choice-making process to complete complex objectives.

‍

Python

An interpreted, advanced, versatile coding language developed by Guido van Rossum and initially launched in 1991. Python's core design principles prioritize easy-to-read syntax, notably incorporating meaningful indentation. Its structural components and object-oriented methodology are intended to assist developers in crafting concise, well-organized code for minor and extensive projects.

‍

PyTorch

A computational learning framework built upon the Torch toolkit, utilized for tasks like image analysis and text interpretation, initially created by Meta AI and currently under the Linux Foundation's jurisdiction.

Q-Learning

A reinforcement learning method without a model that determines the worth of a move in a specific condition.

‍

Qualification Problem

In philosophy and artificial intelligence, particularly in knowledge-based systems, the qualification problem addresses the challenge of enumerating all the necessary preconditions for a real-world action to produce its desired outcome. It can be framed as the question of how to handle the factors that hinder the attainment of my intended result. This problem is closely related to and contrasts with the ramifications aspect of the frame problem.

‍

Quantifier

In logic, quantification defines the number of elements in the domain of discourse that meet the conditions of an open formula. The two primary quantifiers are "for every" and "there is." For instance, in mathematics, quantifiers enable one to express that natural numbers continue indefinitely by stating that for every n, where n is a natural number, there is another number that is one greater than n.

‍

Quantum Computing

Quantum computing involves leveraging quantum-mechanical principles like entanglement and superposition to carry out computations. Quantum machine learning applies these algorithms on quantum computers to accelerate processing, as it operates significantly faster than traditional ML programs and classical computers.

‍

Query Language

Query languages or data query languages (DQLs) are programming languages designed to perform queries in databases and information systems. In general, query languages can be categorized based on whether they are database query languages or information retrieval query languages. The distinction lies in that a database query language seeks to provide accurate answers to specific questions, while an information retrieval query language aims to locate documents that contain information pertinent to a particular subject of interest.

‍

Question and Answer (Q&A)

An AI methodology that allows users to pose inquiries in natural, conversational language and obtain accurate responses. With the emergence of large language models, question-answering has advanced to allow users to ask questions in everyday speech and leverage Retrieval Augmented Generation (RAG) techniques to construct a comprehensive answer from relevant text excerpts extracted from the target document or dataset.

Radial Basis Function Network

In the domain of mathematical modeling, a radial basis function network is an artificial neural network that utilizes radial basis functions as its activation functions. The network's output is a weighted sum of the radial basis functions of the inputs and neuron parameters. Radial basis function networks serve various purposes, such as function approximation, time series forecasting, classification, and system regulation. They were initially proposed in a 1988 publication by Broomhead and Lowe, researchers at the Royal Signals and Radar Establishment.

‍

Random Forest

A guided ML technique that constructs and merges multiple decision trees to form a forest. It is applied to classification and regression tasks in programming languages like R and Python.

‍

Reasoning System

In information technology, an inference system is a software framework that derives conclusions from existing knowledge by employing logical methodologies such as deduction and induction. Inference systems are essential for the development of artificial intelligence and knowledge-based systems.

‍

Recall

For a given set of outcomes from an analyzed document, recall represents the percentage that reflects how many of the expected correct results have been successfully retrieved. This metric applies to various predictive AI applications, including search, classification, and entity identification.

For instance, consider an application designed to detect all dog breeds mentioned in a document. If the document contains references to 10 different dog breeds but the system only retrieves five, all of which are accurate, then the recall rate of the system would be 50%.

‍

Recurrent Neural Networks (RNN)

A neural network architecture frequently employed in NLP and speech recognition that allows prior outputs to be fed back as inputs.

‍

Regression Analysis

A collection of statistical methods for evaluating the associations between a dependent variable, also referred to as the outcome, response variable, or label in machine learning, and one or more independent variables that are assumed to be free of error, commonly known as regressors, predictors, covariates, explanatory variables, or features. The most prevalent type of regression analysis is linear regression, which involves determining the line, or a more intricate linear combination, that best approximates the data based on a defined mathematical criterion.

‍

Regularization

A collection of methods, including dropout, early stopping, and L1 and L2 regularization, aimed at mitigating overfitting and underfitting during the training process of a learning model.

‍

Reinforcement Learning

Reinforcement learning is an ML approach where an algorithm acquires knowledge by engaging with its environment and receives rewards or penalties depending on its actions.

‍

Reinforcement Learning With Human Feedback (RLHF)

A machine learning algorithm that acquires the ability to execute a task by obtaining guidance from human feedback.

‍

Relations

Recognizing associations is a sophisticated NLP function that reveals how components of a statement are interconnected. For instance, in the sentence "Paul is Joan's father," the system will identify that Paul and Joan share a connection, and this data point will include a link attribute designating the relationship as "family" or "parent-child."

‍

Reservoir Computing

A computational paradigm that can be regarded as a generalization of neural networks. Generally, an input signal is introduced into a predetermined, random, dynamic system known as a reservoir, where the system's internal dynamics transform the input into a higher-dimensional representation. A straightforward output mechanism is then trained to interpret the reservoir's state and generate the desired response. The key advantage is that only the output layer undergoes training while the reservoir remains unchanged. Liquid-state machines and echo-state networks are two primary variants of reservoir computing.

‍

Resource Description Framework (RDF)

A set of standards from the World Wide Web Consortium (W3C) initially created as a metadata representation framework. It has evolved into a broad approach for conceptualizing or structuring information within web-based resources, utilizing diverse syntax conventions and data serialization methods. Additionally, it is applied in knowledge management systems.

‍

Responsible AI

Ethical AI is a comprehensive concept that covers the business and moral considerations involved in how organizations implement and utilize AI technologies. In general, responsible AI aims to guarantee that AI systems are transparent, interpretable, equitable, and sustainable.

‍

Restricted Boltzmann Machine (RBM)

A probabilistic, generative artificial neural framework capable of acquiring a likelihood distribution across its collection of inputs.

‍

Rete Algorithm

A template-recognition method for executing logic-based frameworks. This technique was devised to effectively process numerous rules or templates against multiple entities or assertions within a repository of knowledge. It functions to identify which of the system's governing principles should be triggered based on its dataset and stored assertions.

‍

Retrieval-Augmented Generation (RAG)

Retrieval-augmented generation is an AI approach designed to increase the quality of responses generated by large language models by incorporating verified external knowledge sources beyond the original training dataset. Integrating RAG into an LLM-powered question-answering system offers several advantages: 1. ensuring the model has access to up-to-date, credible information, 2. minimizing the occurrence of hallucinations, and 3. allowing source attribution to bolster user confidence in the generated output.

‍

ROAI

Return on artificial intelligence is a shorthand for measuring the return on investment (ROI) associated with a specific AI initiative or expenditure.

‍

Robotics

A cross-disciplinary field of science and engineering encompassing mechanical engineering, electrical engineering, information technology, computer science, and related domains. Robotics focuses on the creation, assembly, functionality, and application of automated machines, along with the development of computational systems for their regulation, sensory input processing, and data management.

‍

Robots

Robots are physical, mechanical devices designed to carry out tasks on behalf of humans. Bots are usually software-based agents that execute functions within a digital application. Bots are sometimes referred to as conversational agents. Both robots and bots can incorporate AI, including machine learning, but they are not required to do so. AI enables robots and bots to complete tasks in more dynamic and sophisticated ways.

‍

R Programming Language

A coding language and open-source software platform for statistical analysis and visualization maintained by the R Foundation for Statistical Computing. The R programming language is extensively utilized by statisticians and data analysts to craft statistical tools and conduct data exploration.

‍

Rule-Based System

In computing, a system based on predefined rules is utilized to store and process knowledge and analyze information in a meaningful manner. It is commonly employed in AI applications and studies. Typically, a rule-based system refers to frameworks that incorporate manually designed or curated sets of rules. Systems developed through automated rule derivation, such as rule-based ML, are generally not classified under this category.

‍

Rules-Based Machine Translation (RBMT)

Recognized as the traditional method of machine translation, it relies on linguistic knowledge of both the source and target languages, allowing words to convey different meanings based on their contextual usage.

SAO (Subject-Action-Object)

Subject-Action-Object (SAO) is an NLP capability that determines the logical roles of sentence components by identifying the entity performing an action, the action being executed, the entity affected by the action, and any additional modifiers or adjuncts present.

‍

Satisfiability

In mathematical logic, satisfiability and validity are basic ideas in semantics. A formula is satisfiable if it is possible to find an interpretation or model that renders the formula true. A formula is valid if every interpretation makes the formula true. The counterparts of these concepts are unsatisfiability and invalidity; specifically, a formula is unsatisfiable if no interpretation makes the formula true, and invalid if there exists an interpretation that makes the formula false. These four concepts are connected in a way that closely mirrors Aristotle's square of opposition.

‍

Search Algorithm

Any algorithm that addresses the search problem, specifically retrieving information stored within a data structure or computed in the search space of a problem domain, whether with discrete or continuous values.

‍

Selection

The phase of a genetic algorithm where individual genomes are selected from a population for subsequent reproduction using a crossover operator.

‍

Selective Linear Definite Clause Resolution

Also known as SLD resolution, it is the fundamental inference rule used in logic programming. It is an enhancement of resolution, which is both sound and refutation complete for Horn clauses.

‍

Self-Attention Mechanism

These processes, also known as attention mechanisms, assist systems in identifying the key elements of input through various methods. They come in multiple forms and are influenced by the way humans can focus on significant details in their surroundings, interpret ambiguity, and encode information.

‍

Self-Management

The process by which computer systems control their own functioning without human involvement.

‍

Self-Supervised Learning

An ML approach where labeled data is generated from the data itself, without depending on past outcome data or external human annotators for labeling or feedback.

‍

Semantic Network

A type of knowledge representation utilized in various NLP applications, where concepts are linked to one another through semantic associations.

‍

Semantic Query

Supports queries and analysis of associative and contextual types. Semantic queries allow for the retrieval of both explicitly and implicitly derived information based on syntactic, semantic, and structural details within the data. They are intended to provide accurate results, possibly the specific selection of a single piece of information, or to address broader, less defined questions through pattern recognition and digital reasoning.

‍

Semantic Reasoner

Also known as a reasoning engine, rules engine, or simply a reasoner, it is a software tool capable of deducing logical consequences from a set of asserted facts or axioms. The concept of a semantic reasoner extends the idea of an inference engine by offering a more comprehensive set of mechanisms. Inference rules are typically defined using an ontology language, often a description logic language. Many reasoners employ first-order predicate logic for reasoning, with inference typically occurring through forward chaining and backward chaining.

‍

Semantics

Semantics is the examination of the meanings conveyed by words and sentences. It focuses on the connection between linguistic structures and non-linguistic concepts, as well as cognitive representations, to clarify how sentences are interpreted by individuals who speak a language.

‍

Semantic Search

The application of natural language technologies to improve user search functionality by analyzing the connections and implicit meaning between words, uncovering concepts and entities like individuals and organizations, along with their characteristics and interrelations. 

‍

Semi-Structured Data

Data that is organized in a certain manner but does not conform to the tabular format of traditional databases or standard data tables, which are typically arranged in rows and columns. The characteristics of the data vary, even when grouped together. A basic example is a form, while a more complex example is an object-oriented database, where information is structured as interconnected objects.

‍

Sensor Fusion

The integration of sensory data or information obtained from various sources in a way that the resulting data has less uncertainty than would be achievable when these sources are used separately.

‍

Sentiment

Sentiment refers to the overall attitude or emotional tone conveyed within a piece of text.

‍

Sentiment Analysis

Also referred to as opinion mining, sentiment analysis is the use of AI to evaluate the tone and sentiment expressed in a given text.

‍

Separation Logic

An expansion of Hoare logic, a method for reasoning about programs. The assertion language of separation logic is a specific instance of the logic of bunched implications (BI).

‍

Similarity and Correlation

Similarity is a natural language processing function that identifies and retrieves documents resembling a specified document. It typically assigns a score to represent the degree of resemblance between each retrieved document and the reference document in the query. However, there is no universally accepted method for assessing similarity, meaning this evaluation is often tailored to a specific application rather than being a standardized or industry-wide metric.

‍

Similarity Learning

A field of supervised learning closely associated with classification and regression but with the aim of learning from a similarity function that evaluates how similar or connected two objects are. It has uses in ranking, recommendation systems, visual identity tracking, face verification, and speaker verification.

‍

Simple Knowledge Organization System (SKOS)

A standardized data framework for structuring knowledge organization systems, including thesauri, classification frameworks, subject heading schemes, and taxonomies.

‍

Simulated Annealing (SA)

A probabilistic method for estimating the global optimum of a given function. Specifically, it is a metaheuristic used to approximate global optimization in an extensive search space for an optimization problem.

‍

Situated Approach

In artificial intelligence research, the situated approach develops agents intended to act efficiently and successfully within their environment. This approach involves creating AI from the ground up by focusing on the fundamental perceptual and motor abilities necessary for survival. The situated approach places significantly less emphasis on abstract reasoning or problem-solving capabilities.

‍

Situation Calculus

A logical framework created for representing and reasoning about dynamic domains.

‍

Software

A set of data or computer instructions that direct the computer on how to operate. Software contrasts with physical hardware, which forms the system and actually carries out the tasks. Computer software encompasses all information handled by computer systems, including programs and data. It includes computer programs, libraries, and associated non-executable data, such as online documentation or digital media.

‍

Software Engineering

The use of engineering principles in the structured development of software.

‍

SPARQL

A Resource Description Framework (RDF) query language, meaning a semantic query language for databases capable of retrieving and modifying data stored in RDF format.

‍

Sparse Dictionary Learning

Also known as sparse coding or SDL, a feature learning technique focused on discovering a sparse representation of the input data as a linear combination of fundamental components, along with those components themselves.

‍

Spatial-Temporal Reasoning

A field of AI that draws from computer science, cognitive science, and cognitive psychology. The theoretical objective on the cognitive side focuses on representing and reasoning about spatial-temporal knowledge in the mind. The practical aim on the computing side involves creating advanced control systems for automata to navigate and comprehend time and space.

‍

Specialized Corpora

A targeted compilation of data or instructional material utilized to train an AI system. Specialized datasets concentrate on a particular sector, such as banking, insurance, or healthcare, or on a distinct business domain or application, like legal documentation.

‍

Speech Analytics

The procedure of examining recorded or real-time conversations using speech recognition technology to extract valuable insights and ensure quality control. Speech analytics tools recognize spoken words and evaluate audio patterns to identify emotions and levels of stress in a speaker's voice.

‍

Speech Recognition

Speech recognition, also known as automatic speech recognition (ASR), computer speech processing, or speech-to-text, allows a software application to convert spoken language into a textual format.

‍

Spiking Neural Network (SNN)

An artificial neural network that more closely resembles a natural neural network. In addition to the state of neurons and synapses, SNNs integrate the element of time into their operating model.

‍

Stanford Research Institute Problem Solver (STRIPS)

An automated planner created by Nils Nilsson and Richard Fikes at SRI International in 1971.

‍

State

A program is considered stateful if it is built to retain previous events or user interactions; the retained information is referred to as the system's state.

‍

State–Action–Reward–State–Action (SARSA)

A reinforcement learning algorithm for learning a policy in a Markov decision process.

‍

Statistical Classification

In machine learning and statistics, classification is the task of determining which category or sub-population a new observation belongs to based on a training dataset of observations or instances where the category membership is already known. Examples include categorizing an email as spam or non-spam and assigning a diagnosis to a patient based on observed traits such as sex, blood pressure, and the presence or absence of specific symptoms. Classification is a type of pattern recognition.

‍

Statistical Relational Learning (SRL)

A branch of artificial intelligence and machine learning that focuses on domain models exhibiting both uncertainty, which can be handled using statistical techniques, and complex, relational structures. Note that SRL is occasionally referred to as relational machine learning (RML) in academic literature. Typically, the knowledge representation frameworks developed in SRL use a subset of first-order logic to describe the relational properties of a domain in a general way, known as universal quantification, and rely on probabilistic graphical models, such as Bayesian networks or Markov networks, to represent uncertainty; some also draw from the methods of inductive logic programming.

‍

Stochastic Optimization (SO)

Any optimization technique that creates and utilizes random variables. In stochastic problems, these random variables are present in the formulation of the optimization problem itself, involving random objective functions or constraints. Stochastic optimization techniques also encompass methods with random iterations. Some stochastic optimization approaches use random iterations to tackle stochastic problems, blending both interpretations of stochastic optimization. These methods extend deterministic techniques to address deterministic problems.

‍

Stochastic Semantic Analysis

A method employed in computer science as a semantic element of natural language comprehension. Stochastic models typically utilize the definition of word segments as fundamental semantic units for the semantic frameworks and, in certain cases, incorporate a two-tiered approach.

‍

Structured Data

Structured data refers to information that is organized and easily searchable. Examples include phone numbers, dates, and product SKUs.

‍

Subject-Matter Expert (SME)

An individual who has amassed extensive expertise in a specific domain or subject matter, evidenced by their academic qualifications, certification, and/or prolonged years of hands-on professional involvement with the discipline.

‍

Superintelligence

A theoretical entity endowed with cognitive capabilities vastly exceeding those of the most brilliant and highly skilled human intellects. Superintelligence can also denote an attribute of advanced problem-solving frameworks, regardless of whether these exceptional intellectual abilities are integrated into autonomous entities operating within the physical realm. A superintelligence might or might not arise from an intelligence explosion and be linked to a technological singularity.

‍

Supervised Learning

Supervised learning is an ML technique where labeled output data is used to train the system and develop accurate algorithms. It is significantly more prevalent than unsupervised learning.

‍

Summarization (Text) 

Text summarization refers to the technique of generating a concise, precise, and coherent summary of a lengthier text document. The objective is to condense the content while retaining its key information and core meaning. There are two primary approaches to text summarization: extractive summarization and generative summarization, also known as abstractive summarization. 

‍

Support Vector Machines

In machine learning, support vector machines (SVMs), also known as support vector networks, are supervised learning models with corresponding learning algorithms that examine data for classification and regression tasks.

‍

Swarm Intelligence (SI)

The combined behavior of decentralized, self-regulating systems, whether natural or synthetic. The term was first introduced in the context of cellular robotic systems.

‍

Symbolic Artificial Intelligence

The designation for the aggregation of all techniques in artificial intelligence exploration that rely on high-level, symbolic depictions of challenges, reasoning, and exploration.

‍

Symbolic Methodology

A symbolic methodology is a strategy for constructing AI systems for natural language processing that relies on a deterministic, conditional framework. Put differently, a symbolic approach structures a system using precise, narrowly defined instructions that ensure the identification of linguistic patterns. Rule-based solutions typically exhibit a high level of accuracy, though they may demand more effort than machine learning-based approaches to comprehensively address a given problem, depending on the specific application.

‍

Syntax

The sequencing of words and expressions in a particular structure to convey meaning in language. Altering the placement of a single word can potentially modify the context and interpretation.

‍

Synthetic Intelligence (SI)

A substitute term for artificial intelligence that highlights the idea that machine intelligence does not have to be a replica or in any way unnatural; it can be a true form of intelligence.

‍

Systems Neuroscience

A branch of neuroscience and systems biology that investigates the structure and function of neural circuits and systems. It is a broad term that includes several fields focused on how nerve cells interact when linked to form neural pathways, circuits, and larger brain networks.

Taxonomy

A taxonomy is a predefined categorization of a particular knowledge domain. It establishes hierarchical connections between elements using "part of" or "type of" relationships, forming a multi-tiered, branching structure where the terminal node of each branch is called a leaf. This framework imposes organization and hierarchy within knowledge subsets.

Organizations utilize taxonomies to systematically classify their documents, facilitating more efficient retrieval by both internal and external users. These classifications can be unique to a single company or evolve into widely accepted standards shared between industries.

‍

Technological Singularity

A theoretical moment in the future when technological progress becomes unwieldy and unstoppable, leading to incomprehensible transformations in society.

‍

Temperature

A variable that regulates the level of randomness or variability in the output of a large language model. A higher setting results in more diverse and unpredictable responses, while a lower setting ensures the output remains more consistent and deterministic.

‍

Temporal Difference Learning

A category of model-free reinforcement learning techniques that learn by bootstrapping from the present approximation of the value function. These approaches sample from the environment, similar to Monte Carlo methods, and are updated based on current estimates, akin to dynamic programming methods.

‍

TensorFlow

A free and open-source software library for dataflow and differentiable programming for various tasks. It serves as a symbolic mathematics library and is also utilized for ML purposes, such as neural networks.

‍

Tensor Network Theory

A theory of brain activity, especially in the cerebellum, that offers a mathematical model for the conversion of sensory space-time coordinates into motor coordinates and vice versa through cerebellar neural networks. The theory was formulated as a geometrization of brain function, particularly within the central nervous system, using tensors.

‍

Text Analytics

Methods employed to analyze vast amounts of unstructured text or text lacking a predefined, organized format to uncover insights, patterns, and comprehension. This process may involve identifying and categorizing text topics, generating summaries, extracting essential entities, and assessing the tone or sentiment conveyed in the text.

‍

Test Set

A test dataset is a compilation of example documents that reflect the difficulties and varieties of content an ML solution will encounter in a real-world deployment. It is utilized to evaluate the precision of an ML system following a phase of training.

‍

Theoretical Computer Science (TCS)

A branch of general computer science and mathematics that concentrates on the more mathematical aspects of computing, including the theory of computation.

‍

Theory of Computation

The theory of computation is the area that focuses on how effectively problems can be solved on a computational model employing an algorithm. The discipline is divided into three main areas: automata theory and languages, computability theory, and computational complexity theory, all connected by the central question: What are the core abilities and constraints of computers?

‍

Thesauri

A linguistic or terminological reference lexicon that defines connections between words and phrases in a structured representation of natural language, facilitating the utilization of definitions and relationships in text analysis.

‍

Thompson Sampling

A heuristic for selecting actions that tackles the exploration-exploitation trade-off in the multi-armed bandit problem. It involves choosing the action that maximizes the anticipated reward based on a randomly sampled belief.

‍

Time Complexity

The computational complexity that characterizes the duration required to execute an algorithm. Time complexity is typically assessed by tallying the number of basic operations executed by the algorithm, assuming each basic operation requires a constant amount of time. Therefore, the time taken and the number of basic operations performed by the algorithm are considered to differ by, at most, a constant factor.

‍

Token

A token is a fundamental text unit that a large language model utilizes to comprehend and produce language. It can be a complete word or a segment of a word.

‍

Training Data

Training data consists of the information or examples provided to an AI system to help it learn, identify patterns, and generate new content.

‍

Training Set

A training dataset is a pre-labeled collection of sample information provided to a machine learning algorithm, allowing it to understand a problem, identify patterns, and ultimately generate a model capable of detecting those same patterns in future evaluations.

‍

Transfer Learning

Transfer learning is a machine learning technique that utilizes previously acquired knowledge and applies it to new tasks and functions.

‍

Transformer

A kind of deep learning framework that utilizes a multi-head attention mechanism. Transformers overcome certain constraints of long short-term memory and have gained widespread adoption in natural language processing, though they can also handle other data types, such as images, in the case of vision transformers.

‍

Transformer Models

Employed in generative AI, where the T represents transformer, transformer architectures are a category of language models. These are neural networks and fall under the classification of deep learning frameworks. They empower AI systems to identify and concentrate on crucial aspects of both input and output by utilizing a technique known as the self-attention mechanism for assistance.

‍

Transhumanism

Shortened to H+ or h+, a global philosophical movement that promotes altering the human state by advancing and broadly distributing advanced technologies to significantly augment human cognition and physical capabilities.

‍

Transition System

In theoretical computing, a transition system is a notion utilized in the analysis of computation. It serves to represent the possible behaviors of discrete systems. This system comprises states and transitions between them, which may be annotated with labels selected from a predefined set; a single label can be associated with multiple transitions. If the label set contains only one element, the system is effectively unannotated, allowing for a simplified definition that excludes labels.

‍

Treemap

Treemaps represent extensive hierarchically arranged, tree-structured data. The visualization space is divided into rectangles, which are sized and organized based on a numerical variable. The hierarchical levels within the treemap are depicted as larger rectangles enclosing smaller nested rectangles.

‍

Tree Traversal

A method of navigating a graph, referring to the procedure of accessing, inspecting, and/or modifying each vertex within a tree data structure precisely once. These traversals are categorized based on the sequence in which the nodes are explored.

‍

Triple or Triplet Relations

A sophisticated extraction method that detects three components—subject, predicate, and object—that can be utilized for storing information.

‍

True Quantified Boolean Formula

Within computational complexity theory, the language TQBF is a formal linguistic construct composed of valid quantified Boolean expressions. A completely quantified Boolean formula is an expression in quantified propositional logic where each variable is bound by either an existential or universal quantifier at the start of the statement. Such a formula always evaluates to either true or false, as it contains no free variables. If the formula evaluates as true, it belongs to the TQBF language. This concept is also referred to as Quantified SAT (QSAT).

‍

Tunable

An AI model that can be readily tailored to meet particular needs. For instance, it can be adapted for specific industries like healthcare, oil and gas, or for departmental functions such as accounting or human resources.

‍

Tuning

The process of refining a previously trained language model with proprietary data. The original model's parameters are adjusted to reflect the unique attributes of the domain-specific data and the particular task being modeled. This adaptation ensures the most precise results and most meaningful insights.

‍

Turing Machine

A computational mathematical model representing a theoretical machine that processes symbols on a tape-like strip based on a set of predefined rules. Despite its straightforward design, this model can execute any algorithm.

‍

Turing Test

The Turing test, developed by computer scientist Alan Turing, assesses a machine's capability to demonstrate human-like intelligence, particularly in language and behavior. During the test, a human evaluator analyzes conversations between a person and a machine. If the evaluator cannot differentiate between their responses, the machine passes the Turing test.

‍

Type System

In programming languages, a collection of principles that assigns a characteristic known as a type to various elements within a software program, including variables, expressions, functions, and modules. These types establish and enforce the implicit classifications that developers use for algebraic data types, data structures, or other components. The primary goal of a type system is to minimize the likelihood of errors in software applications by defining boundaries between different segments of a program and verifying that these segments interact in a logically consistent manner. This validation can occur statically during compilation, dynamically while executing, or through a blend of both approaches. Type systems also serve additional functions, such as representing business constraints, facilitating specific compiler optimizations, enabling multiple dispatch, acting as a form of documentation, and more.

Unstructured Data 

Unstructured data refers to information that lacks a predefined format and is challenging to search. It includes audio, images, and video content. The majority of data worldwide falls into this category.

‍

Unsupervised Learning

Unsupervised learning is a machine learning approach where an algorithm is trained on unclassified and unlabeled data, allowing it to operate without direct supervision.

‍

User Experience Design/User Interface Design (UX/UI)

User experience (UX) and user interface (UI) design relate to the overall interaction and engagement users have with a product. These methods are not exclusive to AI development. Product designers apply UX/UI principles to create and analyze how users interact with their technologies.

Value-Alignment Complete

A category of microprocessor engineered to expedite machine vision operations.

‍

Vision Processing Unit (VPU)

Comparable to an AI-complete challenge, a value-alignment complete issue is one where resolving the AI control dilemma in its entirety is necessary for its solution.

‍

Voice Recognition

Voice recognition, also known as speech recognition, is a form of human-computer interaction where computers process and interpret spoken language to generate human-like text or audio responses. Examples include Apple's Siri and Amazon's Alexa, which allow for hands-free commands and tasks.

Watson

A query-response computing system designed to provide answers to inquiries expressed in everyday language, created within IBM's DeepQA initiative by a research group headed by lead investigator David Ferrucci. Watson was named in honor of IBM's inaugural CEO, business magnate Thomas J. Watson.

‍

Weak AI

Machine intelligence dedicated to a single specialized function.

‍

Windowing

A technique that utilizes a segment of a document as metadata or contextual reference.

‍

Word Embedding

A depiction of a term in natural language processing. Commonly, this depiction takes the form of a real-valued vector that captures the semantic essence of the word, ensuring that terms positioned nearer in the vector space are presumed to have analogous meanings.

XGBoost

An abbreviation for eXtreme Gradient Boosting, XGBoost is a publicly available software library that offers a penalized gradient boosting framework compatible with various programming languages.

Zero-Shot Extraction

The capability to derive information from text without prior training or predefined labels.