Towards Feedback-to-Plan Decisions for Self-Evolving LLM Agents in CUDA Kernel Generation

Yee Hin Chong, Jiaming Wu, Youhui Zhang, Peng Qu
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:20365-20397, 2026.

Abstract

Large language models (LLMs) have shown strong empirical gains as self-evolving agents for CUDA kernel generation, driven by feedback-conditioned planning across generations. However, how planning decisions attribute and combine heterogeneous feedback signals remains opaque. Standard end-to-end ablations fail to resolve this question, as iterative planning amplifies early perturbations and conflates feedback effects with trajectory-dependent drift. We introduce CUDAnalyst, a unified analysis layer for controlled, generation-level attribution of planning decisions to feedback components via trajectory freezing and selective feedback injection. CUDAnalyst enables stable generation-level evaluation and principled coalitional-style attribution of feedback effects and interactions. Our results show that explicit planning is beneficial only when feedback is aligned, that effective planning emerges from structured multi-feedback interactions, and that high-level plans from stronger reasoning models can partially transfer to weaker ones. These trends hold across reference backbones, representative workloads, and reference induction regimes, indicating that the identified feedback-to-plan structure is robust within the controlled axes studied.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-chong26b, title = {Towards Feedback-to-Plan Decisions for Self-Evolving {LLM} Agents in {CUDA} Kernel Generation}, author = {Chong, Yee Hin and Wu, Jiaming and Zhang, Youhui and Qu, Peng}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {20365--20397}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/chong26b/chong26b.pdf}, url = {https://proceedings.mlr.press/v306/chong26b.html}, abstract = {Large language models (LLMs) have shown strong empirical gains as self-evolving agents for CUDA kernel generation, driven by feedback-conditioned planning across generations. However, how planning decisions attribute and combine heterogeneous feedback signals remains opaque. Standard end-to-end ablations fail to resolve this question, as iterative planning amplifies early perturbations and conflates feedback effects with trajectory-dependent drift. We introduce CUDAnalyst, a unified analysis layer for controlled, generation-level attribution of planning decisions to feedback components via trajectory freezing and selective feedback injection. CUDAnalyst enables stable generation-level evaluation and principled coalitional-style attribution of feedback effects and interactions. Our results show that explicit planning is beneficial only when feedback is aligned, that effective planning emerges from structured multi-feedback interactions, and that high-level plans from stronger reasoning models can partially transfer to weaker ones. These trends hold across reference backbones, representative workloads, and reference induction regimes, indicating that the identified feedback-to-plan structure is robust within the controlled axes studied.} }
Endnote
%0 Conference Paper %T Towards Feedback-to-Plan Decisions for Self-Evolving LLM Agents in CUDA Kernel Generation %A Yee Hin Chong %A Jiaming Wu %A Youhui Zhang %A Peng Qu %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-chong26b %I PMLR %P 20365--20397 %U https://proceedings.mlr.press/v306/chong26b.html %V 306 %X Large language models (LLMs) have shown strong empirical gains as self-evolving agents for CUDA kernel generation, driven by feedback-conditioned planning across generations. However, how planning decisions attribute and combine heterogeneous feedback signals remains opaque. Standard end-to-end ablations fail to resolve this question, as iterative planning amplifies early perturbations and conflates feedback effects with trajectory-dependent drift. We introduce CUDAnalyst, a unified analysis layer for controlled, generation-level attribution of planning decisions to feedback components via trajectory freezing and selective feedback injection. CUDAnalyst enables stable generation-level evaluation and principled coalitional-style attribution of feedback effects and interactions. Our results show that explicit planning is beneficial only when feedback is aligned, that effective planning emerges from structured multi-feedback interactions, and that high-level plans from stronger reasoning models can partially transfer to weaker ones. These trends hold across reference backbones, representative workloads, and reference induction regimes, indicating that the identified feedback-to-plan structure is robust within the controlled axes studied.
APA
Chong, Y.H., Wu, J., Zhang, Y. & Qu, P.. (2026). Towards Feedback-to-Plan Decisions for Self-Evolving LLM Agents in CUDA Kernel Generation. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:20365-20397 Available from https://proceedings.mlr.press/v306/chong26b.html.

Related Material