A Sharp Analysis of Model-based Reinforcement Learning with Self-Play

Authors: Qinghua Liu, Tiancheng Yu, Yu Bai, Chi Jin

ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Theoretical In this paper, we present a sharp analysis of model-based self-play algorithms for multi-agent Markov games. We design an algorithm Optimistic Nash Value Iteration (Nash-VI) for two-player zero-sum Markov games that is able to output an ϵ-approximate Nash policy in O(H3SAB/ϵ2) episodes of game playing.
Researcher Affiliation Collaboration 1Princeton University, 2Massachusetts Institute of Technology, 3Salesforce Research.
Pseudocode Yes Algorithm 1 Optimistic Nash Value Iteration (Nash-VI); Algorithm 2 Optimistic Value Iteration with Zero Reward (VI-Zero)
Open Source Code No The paper does not provide any explicit statement about open-sourcing code or a link to a code repository for the described methodology.
Open Datasets No The paper is theoretical and focuses on algorithm design and theoretical guarantees for Markov games defined by abstract parameters (S, A, B, H), rather than empirical evaluation on specific datasets.
Dataset Splits No The paper is theoretical and does not involve empirical data splits for training, validation, or testing.
Hardware Specification No The paper does not mention any specific hardware used for computational work or experiments.
Software Dependencies No The paper does not specify any software dependencies with version numbers.
Experiment Setup No The paper is theoretical, describing algorithms and their guarantees, and does not provide details on experimental setup such as hyperparameters or training configurations for empirical runs.