AI Paper Atlas

Research map / Reinforcement learning / Visual and representation RL

Reinforcement Learning for Machine Learning Engineering Agents

Sherry Yang, Joy He-Yueya, Percy Liang

ICLR 2026 · Reinforcement learning › Visual and representation RL · approach: Scalable RL systems and environments · 12 citations (OpenAlex)

Read the paper

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
RL training of Qwen2.5-3B instead of prompting it with AIDEQwen2.5-3B · MLE-Bench: random-acts-of-pizzatask score: 0.589 -> 0.663Improved
RL training of Qwen2.5-3B instead of prompting it with AIDEQwen2.5-3B · MLE-Bench: tweet-sentiment-extractiontask score: 0.027 -> 0.596Improved
RL-trained Qwen2.5-3B instead of prompting Claude-3.5-SonnetQwen2.5-3B · MLE-Bench: leaf-classificationlog loss: 0.436 -> 0.124Improved
RL-trained Qwen2.5-3B instead of prompting Claude-3.5-SonnetQwen2.5-3B · MLE-Bench: us-patent-phrase-to-phrase-matchingtask score: 0.805 -> 0.527Hurt
RL-trained Qwen2.5-3B instead of prompting Claude-3.5-SonnetQwen2.5-3B · MLE-Bench (12 tasks average)relative improvement: 22%Improved
RL-trained Qwen2.5-3B instead of best GPT-4o scaffoldQwen2.5-3B · MLE-Bench (12 tasks average)relative improvement: 17.7%Improved
run GPT-4o AIDE for 100 hours instead of 24GPT-4o · MLE-Benchtask scoreNo clear effect
add duration-aware gradient weighting (ablation)RL training · random-acts-of-pizzaexecution time of solutionsImproved
add environment instrumentation partial credit (ablation)RL training · MLE-Bench: plant-pathology-2020-fgvc7average task scoreImproved
add explicit self-improvement prompt (ablation)Qwen2.5-3B RL · MLE-Bench (12 tasks average)relative improvement: 8%Improved
add improve-previous-solution prompt (ablation)Qwen2.5-3B RL · MLE-Bench: leaf-classificationlog loss: 0.469 -> 0.124Improved
add improve-previous-solution prompt (ablation)Qwen2.5-3B RL · MLE-Bench: plant-pathology-2020-fgvc7task score: 0.949 -> 0.970Improved

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