AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

LLM4Cov: Execution-Aware Agentic Learning for High-coverage Testbench Generation

Hejia Zhang, Zhongming Yu, Chia-Tung Ho, Haoxing Ren, Brucek Khailany, Jishen Zhao

ICML 2026 · Reinforcement learning › RL from feedback · approach: Scaling RL with verifiable rewards · 0 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 execution-based filtering to SFT data (ablation)SFT dataset · CVDP-ECov Syn Pass (agentic)syntax pass rate (%): 79.5 -> 87.7Improved
add execution-based filtering to SFT data (ablation)SFT dataset · CVDP-ECov Syn Pass (direct inference)syntax pass rate (%): 70.6 -> 83.9Improved
add augmented spec for the teacher (ablation)Teacher · CVDP-ECov Syn Pass (direct inference)syntax pass rate (%): 53.3 -> 71.1Improved
add augmented spec for the teacher (ablation)Teacher · CVDP-ECov Syn Pass (agentic)syntax pass rate (%): 85.1 -> 85.8No clear effect
add Stage 0 SFTQwen3-4B-2507 · CVDP-ECov (agentic)coverage pass rate (%): 28.4 -> 60.2Improved
add Stage 1 after Stage 0LLM4Cov · CVDP-ECov (agentic)coverage pass rate (%): 60.2 -> 67.0Improved
add Stage 2 after Stage 1LLM4Cov · CVDP-ECov (agentic)coverage pass rate (%): 67.0 -> 69.2Improved
add Stage 2 after Stage 1LLM4Cov · CVDP-ECov (direct inference)coverage pass rate (%): 48.2 -> 50.6Improved
add Stage 2 after Stage 1LLM4Cov · AutoEval-ECov (agentic)coverage pass rate (%): 83.6 -> 85.0Improved
increase agentic rounds from R=0 to R=3 (ablation)LLM4Cov Stage 2 · CVDP-ECovcoverage pass rate (%): 48.2 -> 69.2Improved
use worst-state-prioritized intermediate state selection instead of best, uniform, median (ablation)LLM4Cov Stage 1 · CVDP-ECov (agentic)coverage pass rateImproved
train progressively stage by stage instead of naive data augmentation (ablation)LLM4Cov · CVDP-ECov (agentic)coverage pass rateImproved

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