AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

Adaptive Code Watermarking Through Reinforcement Learning

Zhimeng Guo, Huaisheng Zhu, Siyuan Xu, Hangfan Zhang, Teng Xiao, Minhao Cheng

ICML 2026 · Reinforcement learning › RL from feedback · approach: Scaling RL with verifiable rewards

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 process-based advantage A2 (token-level reward) (ablation)CodeTracer · Reward ablationAUROC: 75.11 -> 82.95Improved
add process-based advantage A2 (token-level reward) (ablation)CodeTracer · Reward ablationTPR: 30.29 -> 46.34Improved
add process-based advantage A2 (token-level reward) (ablation)CodeTracer · Reward ablationPass@1: 61.15 -> 60.82No clear effect
add outcome-based advantage A1 (ablation)CodeTracer · Reward ablationAUROC: 79.52 -> 82.95Improved
add outcome-based advantage A1 (ablation)CodeTracer · Reward ablationTPR: 34.91 -> 46.34Improved
add outcome-based advantage A1 (ablation)CodeTracer · Reward ablationPass@1: 60.34 -> 60.82Improved
add supervised fine-tuning initialization before RL (CodeTracer-2 vs RL only CodeTracer-1) (ablation)CodeTracer · HumanEvalPass@1 / AUROC / TPR: Pass@1 62.65 -> 60.82; AUROC 77.71 -> 82.95; TPR 32.32 -> 46.34Mixed
apply watermark model trained on 1.5B directly to 8B LLM (vs WLLM)CodeTracer · OpenCoder-8B-InstructAUROC: 65.90 -> 78.69Improved
CodeTracer vs SWEET under DIPPER paraphrasing attackCodeTracer · HumanEval (DIPPER attack)AUROC: 53.05 -> 58.42Improved
CodeTracer vs SWEET on Java zero-shot transferCodeTracer · HumanEvalPack JavaPass@1: 25.22 -> 31.05Improved
CodeTracer vs SWEET on C++ zero-shot transferCodeTracer · HumanEvalPack C++Pass@1: 44.85 -> 44.39No 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