Fundamentals of Artificial Intelligence - Examination Practice

Vocabulary practice flashcards covering fundamental concepts, definitions, search strategies, formal logic, probabilistic reasoning, planning, machine learning, and reinforcement learning from the Fundamentals of Artificial Intelligence course.

Thomas Hobbes

A seventeenth-century philosopher identified by George B. Dyson in 'Darwin Among the Machines: The Evolution of Global Intelligence' as the patriarch of artificial intelligence.

Logic Theorist

Considered by many to be the first AI program, designed using heuristic searching as a core concept.

Weak AI

An AI paradigm aiming to build machines that act intelligently without taking a philosophical position on whether or not the machines possess real minds or are actually intelligent.

Commutative Production System

A specialized production system in which the specific order of applying a set of applicable rules to the database does not affect the final result.

Fringe (Search)

The collection of generated nodes that are kept in a specific sequence waiting to be expanded and explored during a search algorithm.

Saddle Point

An outcome within the pay-off matrix of a zero-sum game that is simultaneously the minimum in its row and the maximum in its column.

Data

Primitive, verifiable facts represented in any format or medium.

Information

Interpreted data that has been processed into a meaningful context.

Knowledge

A relation among sets of data that is frequently utilized for the deduction of further information.

Physical Symbol System Hypothesis

The principle asserting that a physical symbol system possesses the necessary and sufficient means for general intelligent action, meaning knowledge is structured symbolically and thinking is symbol manipulation.

Semidecidable

A property of entailment in First-Order Logic indicating that algorithms exist to confirm every entailed sentence, but no algorithm can guarantee a negative answer for every non-entailed sentence.

Conceptualization (in First-Order Logic)

The formal representation of a world consisting fundamentally of objects, relations, and functions.

Axiom

A well-formed formula accepted as true without formal proof; in an AI knowledge system, this encompasses domain-specific knowledge rules and user-provided inputs.

Interpretation

A formal mapping between elements of a logical language and corresponding elements within a conceptualization of the world.

Skolemization

A transformation procedure for systematically eliminating existential quantifiers from a first-order formula in prenex normal form by replacing them with constants or Skolem functions.

Prenex Form

A standardized arrangement of a first-order formula consisting of a leading string of quantifiers (the prefix) followed by a completely quantifier-free expression (the matrix).

Subsumption (of Clauses)

A condition in resolution theorem proving where a clause $\left\{L_i\right\}$ subsumes a clause $\left\{M_i\right\}$ if there exists a substitution $s$ such that $\left\{L_i\right\}s \subseteq \left\{M_i\right\}$.

Domain of Discourse

The non-empty set of objects in the universe over which individual variables range in First-Order Logic.

Markov Property (in Bayesian Networks)

The independence assumption stating that there are no direct dependencies present in the modeled system other than those explicitly specified via directed arcs.

Decision Theory

A theoretical framework combining probability and utility theories to evaluate goal desirability and likelihood of attainment, defining an agent as rational if it maximizes expected utility.

Markov Decision Process (MDP)

A mathematical framework for modeling stochastic decision-making in dynamic environments, defined by a set of states, actions, a transition probability function, and a reward function.

Value Function ($V^\pi(s)$)

The expected cumulative return or long-term discounted reward achieved by an agent starting in state $s$ and consistently executing actions according to policy $\pi$.

Discount Factor ($\gamma$)

A parameter in $[0, 1)$ applied in infinite-horizon MDPs to ensure that infinite-horizon expected rewards sum to a finite value, keeping optimization well-defined.

Frame Problem

The computational and representational dilemma in automated planning of efficiently stating all conditions that remain unchanged after an action is executed.

ABSTRIPS

An early hierarchical planning system that reduces search space by abstracting away non-critical preconditions during early search phases.

Goal Stack Planning

The reasoning strategy utilized by the classic STRIPS planner, employing a last-in-first-out stack to sequence pending subgoals and applicable operators.

Mutex Condition

A mutual exclusion relationship identified during planning graph construction between pairs of actions or propositions that cannot be achieved or executed concurrently without contradiction.

Critic (Learning Agent)

The architectural component of a learning agent that evaluates action outcomes against an external, fixed performance standard to supply learning feedback.

Performance Element (Learning Agent)

The component of a learning agent that receives perceptual inputs, selects appropriate actions, and executes decisions in the environment.

Hypothesis ($h$)

In inductive learning, a candidate function derived from training examples of an unknown function $f$ that serves as an approximation of $f$.

Dendrogram

A tree-like diagram that illustrates the hierarchical, nested sequence of clusters formed during agglomerative or divisive hierarchical clustering.

Intra-cluster Cohesion

A cluster validation metric assessing the proximity of data points within a cluster to their common centroid, reflecting cluster compactness.

Inter-cluster Separation

A cluster evaluation criterion that assesses the distance between different cluster centroids, reflecting cluster isolation.

Support Vectors

The critical training sample points situated closest to the decision boundary or separating hyperplane in a Support Vector Machine classifier.

Kernel Trick

A mathematical technique enabling support vector algorithms to operate in a high-dimensional feature space without explicitly calculating coordinates, using only inner dot products.

Q-learning

A model-free reinforcement learning algorithm designed to learn the quality or value of executing specific state-action pairs rather than values of individual states alone.

Generative Adversarial Networks (GANs)

An unsupervised deep learning framework in which two neural networks (a generator and a discriminator) compete against each other in a zero-sum game.