Deep SPI：世界モデルによる安全な方策改善
Deep SPI: Safe Policy Improvement via World Models
世界モデルと表現学習を組み合わせたオンライン強化学習において、方策更新を現在の方策の近傍に制限することで単調改善と収束を保証する理論的枠組みを構築し、その理論に基づくアルゴリズムDeepSPIを提案した論文。
著者: Florent Delgrange, Raphael Avalos, Willem Röpke
分類: cs.LG, cs.AI
原文アブストラクト
Safe policy improvement (SPI) offers theoretical control over policy updates, yet existing guarantees largely concern offline, tabular reinforcement learning (RL). We study SPI in general online settings, when combined with world model and representation learning. We develop a theoretical framework showing that restricting policy updates to a well-defined neighborhood of the current policy ensures monotonic improvement and convergence. This analysis links transition and reward prediction losses to representation quality, yielding online, "deep" analogues of classical SPI theorems from the offline RL literature. Building on these results, we introduce DeepSPI, a principled on-policy algorithm that couples local transition and reward losses with regularised policy updates. On the ALE-57 benchmark, DeepSPI matches or exceeds strong baselines, including PPO and DeepMDPs, while retaining theoretical guarantees.