AI Paper Atlas

Research map / Reinforcement learning / Visual and representation RL

Scalable Online Planning via Reinforcement Learning Fine-Tuning

Arnaud Fickinger, Hengyuan Hu, Brandon Amos, Stuart Russell, Noam Brown

NeurIPS 2021 · Reinforcement learning › Visual and representation RL · approach: RL for combinatorial routing · 5 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
replace single-agent SPARTA with single-agent RL searchQ-learning blueprint · Hanabi normalaverage score: 24.57 -> 24.59Improved
replace multi-agent SPARTA with multi-agent RL searchQ-learning blueprint · Hanabi normalaverage score: 24.61 -> 24.62Improved
replace multi-agent SPARTA with multi-agent RL searchQ-learning blueprint · Hanabi 2 hintsaverage score: 23.67 -> 23.76Improved
replace single-agent SPARTA with single-agent RL searchQ-learning blueprint · Hanabi 2 hintswin rate (%): 25.85 -> 27.85Improved
replace multi-agent SPARTA with multi-agent RL searchQ-learning blueprint · Hanabi 2 hintswin rate (%): 26.87 -> 31.01Improved
add multi-agent RL search over blueprintQ-learning blueprint · Hanabi normalwin rate (%): 63.20 -> 75.93Improved
use joint (multi-agent) instead of single-agent RL searchRL search · Hanabi 2 hintswin rate (%): 27.85 -> 31.01Improved
replace MCTS with RL fine-tuningPPO blueprint · Ms. Pacmancumulative reward: 5820 -> 8080Improved
add RL fine-tuning instead of continuing PPO, 8.10^5 samplesweak PPO blueprint · Ms. Pacmancumulative reward: 1920 -> 5510Improved
add RL fine-tuning instead of continuing PPO, 8.10^5 samplesrandom blueprint · Ms. Pacmancumulative reward: 1280 -> 2730Improved
amortize MCTS tree over 30 steps instead of replanning each step (ablation)MCTS · Ms. Pacmanepisode returnHurt
use refined value instead of blueprint value for last step (infinite horizon) (ablation)RL fine-tuning · Ms. Pacmancumulative rewardNo clear effect

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