Słownik AI
Kompletny słownik sztucznej inteligencji
Multi-Agent Inverse Reinforcement Learning
Extension of IRL where multiple agents simultaneously learn reward functions from experts demonstrating collective behaviors in shared environments.
Collective Reward
Global reward function shared among all agents in a system, optimizing team performance rather than individual gains.
Individual Reward
Reward function specific to each agent, taking into account their personal actions while considering the influence on other agents in the system.
Cooperative Game
Multi-agent scenario where all agents share a common goal and must coordinate their actions to maximize a collective reward.
Competitive Game
Environment where agents have conflicting objectives, each seeking to maximize their own reward at the expense of other agents.
Mixed Game
Multi-agent configuration combining cooperative and competitive elements, where some agents may form temporary coalitions or strategic oppositions.
Preference Alignment
Process of harmonizing individual agents' reward functions to achieve consistency with the global objectives of the multi-agent system.
Multi-Agent Learning from Demonstration
Technique where agents infer rewards from trajectories demonstrated by experts operating simultaneously in the environment.
Nash Equilibrium in IRL
Strategic convergence point where no agent can improve their reward by unilaterally changing their policy, given the policies of other agents.
Joint Value Function
Estimation of the expected cumulative reward for all agents, considering their combined states and actions in a joint state space.
Team Policy
Coordinated strategy defining optimal actions for each agent based on the global state and collective intentions of the system.
Reward Decomposition
Method of separating a global reward into individual components attributable to each agent while preserving collective optimality.
Multi-agent Adversarial Learning
Framework where adversarial agents simultaneously learn to identify and exploit weaknesses in other agents' policies within an IRL context.
Multi-agent Consensus
Process of agreement among agents on a common reward function or shared objectives, necessary for effective cooperative learning.
Implicit Communication
Information transmission between agents through their observable actions and states, without a direct explicit communication channel in the IRL environment.
Federated Learning in IRL
Technique allowing agents to learn rewards from distributed data without sharing their raw data, preserving privacy while collaborating.
Multi-Agent Game Theory
Theoretical framework analyzing strategic interactions between rational agents in inverse reinforcement learning environments.
Transferable Multi-Agent Learning
Ability to transfer learned reward knowledge from one multi-agent context to another, accelerating adaptation to new environments.