1619 terms
Also known as: A star Search · A-star Search
A* Search is a graph search and pathfinding algorithm that uses both the cost already incurred and a heuristic estimate of remaining cost to choose the next path to explore. When its heuristic is properly constrained, it can find an optimal path efficiently and is widely used in planning, navigation, and search problems.
Also known as: A and B Testing · A B Testing
A/B Testing is a controlled statistical comparison of two or more variants of a system, model, interface, or process. It measures whether one variant performs better on selected metrics and whether the observed difference is likely to be meaningful rather than random.
Abductive Logic Programming is a declarative knowledge-representation approach that uses abductive reasoning to infer plausible explanations for observations. It extends ordinary logic programming by allowing selected predicates to remain open as possible assumptions that can be chosen to satisfy a goal.
Abductive Reasoning is a form of inference that starts from observations and seeks the most plausible explanation for them. Unlike deduction, it supports a reasonable hypothesis rather than a conclusively proven result.
Ablation is the deliberate removal of a feature, component, subsystem, or process from an AI or machine learning system to evaluate its contribution. An ablation study compares system behavior before and after removal, often by retraining or retesting, to determine whether the removed element materially affects performance or reliability.
An Abstract Data Type is a mathematical or logical description of a data type based on its externally observable behavior. It specifies the values and operations available to users without requiring knowledge of the underlying implementation.
Abstraction is the practice of suppressing details that are not relevant to a particular analysis, design, or representation. In AI and software systems, abstraction helps focus attention on essential properties while hiding lower-level implementation, physical, temporal, or spatial complexity.
Also known as: XLA
XLA, or Accelerated Linear Algebra, is an open-source compiler technology for optimizing machine learning computations across CPUs, GPUs, and specialized accelerators. It translates model operations from frameworks such as TensorFlow, JAX, or PyTorch into forms intended to execute more efficiently on the target hardware.
Accelerating Change is the view that technological development increases in speed and impact over time. In AI discussions, the term is often used to describe expectations that future technical, economic, and social changes may occur faster than earlier historical patterns.
Also known as: Accessibility (Technology) · Accessibility Technology
Accessibility in technology is the design and provision of digital systems so that people with diverse abilities can use them effectively. It includes inclusive design practices and assistive technologies such as screen readers, speech recognition, captions, and alternative input devices.
Accuracy is a classification metric equal to the proportion of predictions that are correct among all predictions made. It is useful in some evaluation settings but can be misleading when classes are imbalanced, where precision, recall, or other metrics may better reflect model performance.
Act is the stage of an agentic process in which an agent carries out a selected operation. The action may include calling a tool, sending an API request, changing a system state, or otherwise executing the result of a prior reasoning step.
In reinforcement learning, an Action is a choice made by an agent that changes, or attempts to change, the state of the environment. Actions are selected according to a policy and are evaluated through their consequences and rewards.
An Action Language is a formal language used to describe how actions change the state of a system over time. It is used in AI and robotics to model action effects, preconditions, and transitions for planning and reasoning.
Action Model Learning is a machine learning area concerned with learning or revising an agent's knowledge of actions, including their preconditions and effects. The learned model is commonly represented in a formal action language and used by automated planners.
Action Selection is the problem of deciding what an intelligent system should do next. It applies to agents, robots, and cognitive systems that must choose among possible behaviors in response to goals, context, and environmental conditions.
Action Space is the set of tools, operations, permissions, or moves available to an agent when performing a task. A well-scoped action space should be sufficient for the task while limiting unnecessary capabilities that may increase errors or governance risk.
Actionable Intelligence is information that is specific, relevant, and reliable enough to support a decision or intervention. In AI governance and operations, it converts data or analysis into guidance that can be acted upon.
An Activation Function is a function used in a neural network node to transform weighted inputs into an output passed to later layers. By introducing nonlinear behavior, activation functions enable neural networks to model complex relationships rather than only simple linear patterns.
Active Learning is a training approach in which a model or learning system selectively requests the examples or labels that are most useful for improving performance. It is especially valuable when labeled data is scarce, expensive, or dependent on expert human annotation.
AdaGrad is an adaptive gradient-based optimization algorithm that adjusts learning rates separately for individual model parameters. It increases sensitivity to parameter-specific update histories, which can help with sparse features but may reduce learning rates substantially over time.
Adaptable AI describes AI systems that can adjust their behavior, strategies, or internal models in response to changing tasks, environments, or user needs. The concept emphasizes robustness and flexibility without requiring full manual reprogramming for every new condition.
Adaptation is the process of adjusting a model or system to better fit a particular task, domain, dataset, or operational context. In machine learning, it is often used as a synonym for tuning or fine-tuning.
An Adaptive Algorithm is an algorithm that changes its behavior during operation according to defined criteria, feedback, or reward signals. Such algorithms are designed to respond to changing data, environments, or performance conditions rather than follow a fixed execution pattern.
Adaptive Learning refers to educational systems that personalize instruction, feedback, pacing, or content using data about a learner's needs and performance. AI-enabled adaptive learning platforms aim to match learning experiences to individual ability, progress, and context.
An Adaptive Neuro Fuzzy Inference System is a hybrid model that combines neural network learning with fuzzy logic reasoning. It typically uses fuzzy IF-THEN rules with trainable parameters to approximate nonlinear functions and support interpretable decision processes.
An Admissible Heuristic is a heuristic function that never overestimates the true minimum cost of reaching a goal. In pathfinding and search algorithms such as A*, admissibility helps preserve optimality of the solution.
Adversarial Learning is a training approach that exposes models to deliberately challenging, manipulative, or attack-like inputs. Its purpose is to improve robustness, security, and reliability against adversarial behavior or distributional stress.
Also known as: AEO (Answer Engine Optimization) · AEO · Answer Engine Optimization · Answer Engine Optimisation (AEO)
AEO, or Answer Engine Optimization, is the practice of structuring and improving content so that AI-powered answer systems can identify, summarize, and cite it as a useful source. It extends search optimization toward direct-answer environments such as AI search, chat interfaces, voice assistants, and generated overviews.
Affective Computing is the field concerned with systems that recognize, interpret, process, or simulate human emotions and affective states. It draws on computer science, psychology, and cognitive science and may involve signals such as voice, facial expression, text, or interaction behavior.
