Research map / Large language models / Agents, code and math
Towards Feedback-to-Plan Decisions for Self-Evolving LLM Agents in CUDA Kernel Generation
Yee Hin Chong, Jiaming Wu, Youhui Zhang, Peng Qu
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add explicit planning without feedback (P+NF) (ablation) | OpenEvolve CUDA agent · PolyBench-ACC | generation-level success rate | Hurt |
| add aligned feedback to explicit planning (P+F) (ablation) | explicit planning · PolyBench-ACC | generation-level success rate | Improved |
| replace planner output with content-free DummyPlan (ablation) | explicit planning · PolyBench-ACC | generation-level success rate | Hurt |
| randomize feedback across programs (P+RF) (ablation) | explicit planning · PolyBench-ACC | generation-level success rate | Hurt |
| add summarization of feedback (P+S vs P+F) (ablation) | explicit planning, weak models · PolyBench-ACC | overall success | Improved |
| add summarization of feedback (P+S vs P+F) (ablation) | explicit planning, strong models · PolyBench-ACC | overall success | No clear effect |
| add explicit planning over summaries alone (P+S vs NP+S) (ablation) | summarized feedback · PolyBench-ACC | execution success | Improved |
| inject strong-model plans into weak model (ablation) | DeepSeek-V3.2, Qwen3-Coder-30B · PolyBench-ACC | execution success | Improved |
| use same-family plan distillation (ablation) | plan transfer · PolyBench-ACC | execution success | Improved |
| raise attempts k beyond 5 (ablation) | pass@k evaluation · PolyBench-ACC 3DCONV, DeepSeek-V3.2 | pass rate | No clear effect |
| add multi-tool feedback interaction (debugger, analyzer, profiler jointly) | planning · PolyBench-ACC | synergy sigma_FP: 0.233 -> 0.430 | Improved |
Closest papers by meaning
- Language Control Diffusion: Efficiently Scaling through Space, Time, and Tasks (ICLR 2024 · similarity 0.87)
- SPIRAL: Symbolic LLM Planning via Grounded and Reflective Search (AAAI 2026 · similarity 0.87)
- Non-myopic Generation of Language Models for Reasoning and Planning (ICLR 2025 · similarity 0.87)
- LLMs Can Plan Only If We Tell Them (ICLR 2025 · similarity 0.87)
- Agent2World: Learning to Generate Symbolic World Models via Adaptive Multi-Agent Feedback (NeurIPS 2026 · similarity 0.87)
- Beyond Test-Time Memory: State-Space Optimal Control for LLM Reasoning (ICML 2026 · similarity 0.87)
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.
