AI Paper Atlas

Research map / Large language models / Instruction and fine-tuning

ATGen: Adversarial Reinforcement Learning for Test Case Generation

Qingyao Li, Xinyi Dai, Weiwen Liu, Xiangyang Li, Yasheng Wang, Ruiming Tang, Yong Yu, Weinan Zhang

ICLR 2026 · Large language models › Instruction and fine-tuning · approach: Fine-tuning for reasoning data · 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
train test generator with GRPO RL instead of SFT (UTGen)Qwen2.5-7B-Instruct · APPS+Codeforces subsetattack rate (%): 16.24 -> 34.02Improved
add adversarial code generation (Adaptive) (ablation)ATGen 7B · APPS+Codeforces subsetattack rate (%): 34.02 -> 36.99Improved
add adversarial code generation (Unconditional) (ablation)ATGen 3B · APPS+Codeforces subsetattack rate (%): 29.96 -> 32.38Improved
add adversarial code generation (Unconditional) (ablation)ATGen 7B · APPS+Codeforces subsetattack rate (%): 34.02 -> 34.57No clear effect
add adversarial code generation (Adaptive) (ablation)ATGen 7B · APPS+Codeforces subset (Hard)attack rate (%): 18.42 -> 21.05Improved
add adversarial code generation (Adaptive) (ablation)ATGen 7B · APPS+Codeforces subsetIO accuracy (%): 71.56 -> 74.42Improved
reward only the final attack instead of IO acc plus input attack (ablation)ATGen (w/o Adver) 3B · APPS+Codeforces subsetIO accuracy (%): 44.67 -> 67.72Improved
vary reward composition (ablation)ATGen (w/o Adver) 3B · APPS+Codeforces subsetattack rate (%): 30.07 / 29.74 / 30.29No clear effect
increase sampling retries 10 -> 30 (ablation)ATGen 7B · APPS+Codeforces subsetIO accuracy (%): 74.09 -> 70.69Hurt
increase sampling retries 10 -> 30 (ablation)ATGen 7B · APPS+Codeforces subsetinput attack rate (%): 47.96 -> 49.94Improved
use sampling-based instead of instruction-based adversarial code (ablation)ATGen (Adaptive) 7B · APPS+Codeforces subsetattack rate (%): 32.71 -> 36.22Improved
use sampling-based instead of instruction-based adversarial code (ablation)ATGen (Adaptive) 7B · APPS+Codeforces subsetIO accuracy (%): 76.28 -> 74.09Hurt

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