YZ Sözlüğü
Yapay Zekanın tam sözlüğü
Zero-sum game
Competitive situation where the total gain of participants is constant, with one agent's gains exactly corresponding to the losses of other agents.
Nash equilibrium
Game state where no agent can improve their strategy by unilaterally changing their actions, given that other agents' strategies remain fixed.
Minimax theorem
Fundamental principle establishing that in zero-sum games with perfect information, there exists an optimal strategy that maximizes the minimum possible gain.
Game tree
Tree structure representing all possible sequences of moves and resulting states of a game, used for decision analysis.
Alpha-beta pruning
Optimization technique eliminating unnecessary branches of the game tree by maintaining alpha and beta bounds on possible values.
Pure strategy
Deterministic strategy where an agent systematically chooses the same action in a given situation without any randomization.
Mixed strategy
Probabilistic approach where an agent randomly selects among several actions according to a predefined probability distribution.
Payoff matrix
Two-dimensional table representing each agent's payoffs for all possible combinations of players' actions.
Saddle point
Equilibrium position in the payoff matrix where the value is simultaneously the maximum of its column and the minimum of its row.
Duality in game theory
Mathematical relationship between a player's maximization problem and their opponent's minimization problem in zero-sum games.
Backpropagation evaluation
Method of propagating values from the leaves to the root of the game tree to assess the quality of each position.
Iterative deepening search
Algorithm that progressively explores the game tree by increasing the search depth at each iteration to guarantee a limited response time.
Heuristic evaluation function
Function that estimates the value of a game position without completely exploring the tree, used when exhaustive search is impossible.
Value of a state
Quantitative measure representing the optimal expected payoff from a given state assuming perfect play by all agents.
Adversarial learning
Process where agents mutually improve by competing against each other, with each agent developing strategies to counter the others.
Extensive form of the game
Detailed representation of a game including the temporal sequence of decisions, the information available at each stage, and the final payoffs.
Expected Utility Theory
Theoretical framework evaluating decisions based on outcomes weighted by their probabilities in a competitive context.
Perfect Equilibrium Strategy
Strategy constituting a Nash equilibrium for every subgame of the extensive form, ensuring temporal consistency of decisions.
Perfect Information
Condition where each agent knows the complete history of previous actions and the current state of the game before making their decision.
Backward Induction Principle
Method for solving sequential games by analyzing decisions from the end of the game towards the beginning to determine optimal strategies.