Multi-Agent Systems

Agent definitions and properties (autonomy, reactivity, proactivity, social ability), Multi-agent vs. single-agent systems, Agent architectures (reactive, deliberative, hybrid, BDI), Environments (accessible, deterministic, dynamic, continuous, multi-agent), Coordination vs. cooperation vs. competition, Communication languages (KQML, FIPA-ACL), Agent platforms (JADE, SPADE).

Task decomposition and allocation (Contract Net Protocol, auction-based), Centralized vs. distributed planning, Partial global planning and approximate coordination, Robustness to agent failures, Negotiation protocols (monotonic concession, argumentation-based), Game-theoretic negotiation (Nash equilibria, subgame perfect equilibrium), Incentive-compatible mechanisms.

Centralized Training with Decentralized Execution (CTDE), Independent Q-Learning (IQL) limitations, Value Decomposition Networks (VDN), QMIX monotonic factorization, MADDPG for continuous control, Counterfactual Multi-Agent Policy Gradients (COMA), Communication in MARL (differentiable communication channels, gated attention), Credit assignment problem.

Zero-sum and general-sum games, Minimax and alpha-beta pruning, Nash equilibria computation (fictitious play, no-regret learning), Policy gradient methods in games (self-play, population-based training), AlphaGo/AlphaZero algorithms, MuZero model-based planning, OpenAI Five and multi-agent Dota 2, Elo rating systems for agent evaluation.

Swarm robotics coordination (flocking, formation control), Traffic management and autonomous vehicle platooning, Smart grids and energy management MAS, Financial market multi-agent simulation, Human-agent collaboration (shared workspace, mixed-initiative), Hierarchical multi-agent systems, Safe AI and value alignment in MAS, Decentralized AI governance.