Fictitious play — what is it?
hardAnswer
- Iterative: each iteration, each player computes best response to the empirical distribution of opponents' past strategies.
- Converges to Nash in zero-sum games (Robinson 1951).
- Foundation of self-play theory.
- Modern deep variants: NFSP (Neural Fictitious Self-Play, Heinrich & Silver 2016) — combines RL best response with supervised average-strategy learning.
- PSRO (Policy Space Response Oracle) generalizes further.
Check yourself — multiple choice
- Random
- Each iteration, best response to empirical distribution of opponents' past; converges to Nash in zero-sum; NFSP (neural) + PSRO modern extensions
- Same as MCTS
- Not real
Fictitious play: best-respond to opponent history; NFSP / PSRO neural extensions.
#multi-agent#theory
Practise Reinforcement Learning
214 interview questions in this topic.