AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

MM-ReCoder: Advancing Chart-to-Code Generation with Reinforcement Learning and Self-Correction

Zitian Tang, Xu Zhang, Jianbo Yuan, Yang Zou, Varad Gunjal, Songyao Jiang, Davide Modolo

CVPR 2026 · Reinforcement learning › RL from feedback · approach: RL for LLM reasoning · 4 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
add shared-first-turn RL stage before full-trajectory RL (ablation)MM-ReCoder RL · ChartMimic low-level (second turn)low-level score: 83.9 -> 86.0Improved
use shared-first-turn instead of full-trajectory for self-correction (ablation)MM-ReCoder RL · ChartMimicpercentage of improved samples: 3.4 -> 14.4Improved
add self-correction bonus eta=0.1 with gamma=0 in full-trajectory (ablation)MM-ReCoder RL · ChartMimic first turnfirst-turn exec rate: 91.8 -> 64.5Hurt
add model-based reward (ablation)MM-ReCoder RL · ChartMimic single-turnhigh-level score: 78.6 -> 83.7Improved
add rule-based reward (ablation)MM-ReCoder RL · ChartMimic single-turnlow-level score: 78.2 -> 83.9Improved
use Qwen2.5-VL-72B instead of 7B as reward model (ablation)MM-ReCoder RL · ChartMimic single-turnlow-level score: 81.0 -> 84.3Improved
add multi-turn cold start data after single-turn cold start (ablation)MM-ReCoder · ChartMimicsecond-turn low-level: 84.8 -> 86.0Improved
add any cold start versus none (ablation)MM-ReCoder · ChartMimicsecond-turn low-level: 84.9 -> 86.0Mixed
train full-trajectory RL for 2 epochs instead of 1 (ablation)MM-ReCoder RL · ChartMimicsecond-turn low-level: 83.9 -> 81.8Hurt
increase reward weights on model-based reward (ablation)MM-ReCoder RL · ChartMimic single-turnlow-level score: 84.8 -> 83.9No clear effect
add more self-correction turns at inference (2 to 4)MM-ReCoder · ChartMimiclow-level score: 84.8 -> 86.5Improved
add a fifth self-correction turnMM-ReCoder · ChartMimiclow-level score: 86.5 -> 86.4No 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