General AI Terms

• Artificial Intelligence (AI): The study and design of intelligent agents that perceive, reason, learn, and act in an environment.
• Agent: An entity that perceives its environment through sensors and acts upon that environment through actuators.
• Rational Agent: An agent that always selects the action that maximizes its expected performance measure, based on its knowledge.
• Autonomous Agent: An agent that operates without human intervention, continuously and adaptively.
• Environment: Everything external to the agent that it can sense and interact with.
• Percept: An agent’s input from the environment at a given moment.
• Percept Sequence: The complete history of everything an agent has perceived.
• Actuator: The mechanism by which an agent affects the environment (e.g., wheels, arms, output signals).
• Sensor: The mechanism by which an agent receives input from the environment (e.g., cameras, microphones).

Search and Problem Solving

• Search Algorithm: A method for exploring possible states or actions to find a path from a start state to a goal state.
• State Space: A representation of all possible configurations or situations in a search problem.
• Initial State: The starting point in a search or planning problem.
• Goal State: A desired state or set of states that the agent aims to reach.
• Transition Model: Describes how an action changes the state of the environment.
• Path Cost: The cumulative cost of a sequence of actions.
• Uninformed Search: Search algorithms that use no information other than the problem definition (e.g., BFS, DFS, UCS).
• Informed Search: Search algorithms that use domain-specific heuristics to guide the search (e.g., A*).
• Heuristic: An estimate of the cost to reach the goal from a given state.
• Optimality: A property of a search algorithm that guarantees finding the least-cost solution.
• Completeness: A property that ensures the algorithm will find a solution if one exists.

Game and Adversarial Search

• Minimax Algorithm: A strategy for two-player games that maximizes the agent’s minimum possible gain.
• Alpha-Beta Pruning: An optimization of minimax that prunes branches that cannot affect the final decision.
• Utility Function: A numerical representation of the desirability of a state, used in game theory.
• Zero-Sum Game: A game where one player’s gain is another’s loss.

Logic and Knowledge Representation

• Propositional Logic: A formal language where statements are either true or false.
• First-Order Logic (FOL): An extension of propositional logic that includes quantifiers and predicates to express relationships among objects.
• Knowledge Base (KB): A set of logical sentences representing facts and rules.
• Inference: The process of deriving new logical conclusions from known facts.
• Entailment: A relationship where a sentence logically follows from others (e.g., KB ⊨ α).
• Model (Logical): An interpretation that assigns meaning to logical expressions and determines their truth.
• Satisfiability: A KB is satisfiable if there exists at least one model where all sentences are true.
• Resolution: A complete inference technique for propositional logic.

Learning and Machine Learning

• Machine Learning (ML): The study of algorithms that improve automatically through experience.
• Supervised Learning: Learning from labeled examples (input-output pairs).
• Unsupervised Learning: Learning patterns or structure in data without labels.
• Reinforcement Learning (RL): Learning to make sequences of decisions by receiving rewards or penalties from the environment.
• Classifier: An algorithm that assigns a label to input data.
• Neural Network: A computational model inspired by the brain, consisting of layers of nodes (neurons) that learn representations.
• Overfitting: A condition where a model performs well on training data but poorly on unseen data.
• Underfitting: A condition where a model is too simple to capture underlying patterns in the data.
• Loss Function: A function that measures the error between predicted and actual values.
• Gradient Descent: An optimization algorithm used to minimize a loss function by iteratively adjusting parameters.

Probabilistic Reasoning

• Probability Distribution: A function that assigns probabilities to different outcomes.
• Bayesian Network: A directed acyclic graph representing probabilistic dependencies among variables.
• Conditional Probability: The probability of an event given that another event has occurred.
• Markov Decision Process (MDP): A framework for modeling decision-making in fully observable, stochastic environments.
• Partially Observable Markov Decision Process (POMDP): An extension of MDPs where the agent cannot fully observe the environment.

Natural Language Processing (NLP)

• Tokenization: Splitting text into words, phrases, or symbols.
• Parsing: Analyzing the grammatical structure of a sentence.
• Named Entity Recognition (NER): Identifying proper nouns such as names, locations, and dates in text.
• Language Model: A probabilistic model that assigns likelihoods to sequences of words.

Evolutionary Algorithms

• Genetic Algorithm (GA): A search heuristic that mimics the process of natural selection.
• Chromosome: A candidate solution represented as a sequence of genes.
• Gene: A component of a chromosome representing part of a solution.
• Mutation: A random change to an individual’s genes to introduce diversity.
• Crossover: Combining genes from two parents to form offspring.
• Fitness Function: Evaluates how good a solution is for the current problem.
• Selection: The process of choosing individuals from a population to breed the next generation.
• Mixing Number: The number of parents involved in producing an offspring (1 = asexual, 2 = typical crossover, 3+ = multi-parent recombination).
• Stochastic Universal Sampling (SUS): A selection method that places evenly spaced selection points to proportionally sample individuals based on fitness.

Planning and Acting

• Plan: A sequence of actions designed to achieve a goal.
• Planning Agent: An agent that constructs a plan before acting.
• STRIPS: A formal language used for automated planning.
• Effect: The resulting state or outcome of performing an action.

Advanced Learning and Reasoning

• PAC Learning (Probably Approximately Correct): A framework that analyzes whether a learning algorithm can, with high probability, learn a function that is approximately correct given a certain amount of data.
• Inductive Logic Programming (ILP): A technique that combines logic programming with machine learning by inducing logical rules from examples.
• Statistical Relational Learning (SRL): Combines probabilistic reasoning with relational logic to model complex, structured data (e.g., social networks, biological systems).
• Ensemble Learning: Combines multiple models to produce better predictive performance (e.g., random forests, boosting).
• Bayesian Inference: A method of updating probabilities based on evidence using Bayes’ Theorem.
• Variational Inference: An optimization-based method for approximating complex probability distributions, commonly used in deep generative models.
• Expectation-Maximization (EM): An iterative algorithm for finding maximum likelihood estimates in models with hidden variables.

Planning and Acting Under Uncertainty

• Contingency Planning: Creating plans that include branches for different possible future states, useful in uncertain environments.
• Policy: A mapping from states to actions in reinforcement learning or decision processes.
• Value Function: Estimates the expected return (future reward) from a state or state-action pair.
• Bellman Equation: A recursive equation that defines the value of a state in terms of the values of successor states.
• Temporal Difference (TD) Learning: A class of reinforcement learning algorithms that learn predictions based on the difference between successive estimates.
• Q-Learning: An off-policy TD algorithm that learns the expected utility of actions without a model of the environment.
• SARSA (State-Action-Reward-State-Action): An on-policy TD learning algorithm that updates Q-values based on the actual action taken.
• Deep Q-Network (DQN): Combines Q-learning with deep neural networks to approximate value functions in high-dimensional spaces.

Multi-Agent Systems

• Multi-Agent System (MAS): A system composed of multiple interacting intelligent agents, which may cooperate or compete.
• Cooperative Game: A multi-agent setting where all agents share the same reward function.
• Competitive Game (Non-Cooperative): A setting where agents have opposing goals, often modeled as zero-sum.
• Nash Equilibrium: A solution concept where no agent can benefit by changing its strategy while others keep theirs unchanged.
• Joint Policy: A policy that defines a combined strategy for multiple agents acting together.

Advanced Knowledge Representation

• Description Logic: A family of formal languages used to represent structured knowledge and support reasoning, especially in semantic web and ontology systems.
• Ontology: A formal specification of a set of concepts and relationships within a domain.
• Belief-Desire-Intention (BDI) Model: A model for rational agents that distinguishes between beliefs (what is known), desires (goals), and intentions (committed plans).
• Nonmonotonic Reasoning: Reasoning systems where adding new knowledge can invalidate previous conclusions (e.g., default reasoning).
• Default Logic: A formal system for reasoning with defaults or assumptions that hold in the absence of contrary information.

Advanced Probabilistic Models

• Hidden Markov Model (HMM): A statistical model where the system is assumed to be a Markov process with hidden (unobservable) states.
• Kalman Filter: An algorithm that estimates the state of a dynamic system from noisy observations, assuming linearity and Gaussian noise.
• Particle Filter: A nonparametric algorithm that estimates the posterior distribution of state by maintaining a set of samples (particles).
• Factor Graphs: Bipartite graphs representing how a global function factors into a product of local functions; used in probabilistic inference.

Deep Learning & Neural Architectures

• Autoencoder: A neural network trained to reconstruct its input, often used for dimensionality reduction or anomaly detection.
• Variational Autoencoder (VAE): A probabilistic generative model that learns a latent variable representation of input data.
• Generative Adversarial Network (GAN): A generative model where two neural networks (generator and discriminator) compete to produce realistic synthetic data.
• Transformer: A deep learning architecture based on self-attention mechanisms, widely used in NLP (e.g., GPT, BERT).
• Attention Mechanism: Allows a model to focus on specific parts of input data, improving performance in sequence modeling tasks.

Ethics, Safety, and Explainability

• Explainable AI (XAI): Designing AI systems whose decisions can be understood and trusted by humans.
• AI Alignment: Ensuring that AI systems pursue goals that are aligned with human values and intentions.
• Fairness in AI: The study and implementation of algorithms that avoid bias and ensure equitable treatment across demographic groups.
• Adversarial Example: A perturbed input designed to fool a machine learning model into making incorrect predictions.
• Robustness: The ability of a model to maintain performance under noisy, incomplete, or adversarial inputs.