AI Paper Atlas

Research map / Reinforcement learning / Policy gradient and actor-critic

Efficient Exploration via Epistemic-Risk-Seeking Policy Optimization

Brendan O'Donoghue

ICML 2023 · Reinforcement learning › Policy gradient and actor-critic · approach: Model-based RL with guarantees · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Use epistemic-risk-seeking policy optimization instead of vanilla actor-criticDeepSea · DeepSeamaximum depth solved: 14 -> 100Improved
Use epistemic-risk-seeking policy optimization instead of Bootstrapped DQNDeepSea · DeepSeamaximum depth solved: 50 -> 100Improved
Add simple optimism bonuses to actor-criticactor-critic · DeepSeamaximum depth solved: approximately depth 50Improved
Add replay with prioritized sampling and off-policy correctionsERSAC · DeepSeadata efficiency: about 4×Improved
Add Gaussian noise to reward targets in replay (ablation)ERSAC + replay · DeepSeaperformanceImproved
Add replay data (ablation)ERSAC · DeepSea, depth 100seeds reaching goal: 99 times out of 100 -> 93 timesHurt
Learn the risk-seeking parameter τ rather than fix it (ablation)ERSAC · DeepSearobustness across initial τ choicesImproved
Use TD-λ values of at least 0.6 (ablation)ERSAC · DeepSeainstances solved to depth 100: λ ≥ 0.6Improved
Use longer TD-λ rollouts rather than very small rollouts (ablation)ERSAC · DeepSeaexploration performance: 1 -> 25Improved
Add the risk-seeking objective to replay actor-criticAtari 57 · Atari 57frames to reach baseline peak median performance: 1.8× fewerImproved
Add the risk-seeking objective to replay actor-critichard-exploration Atari games · 7 hard-exploration Atari gamesgame performance: four of the 7 gamesMixed

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy