Component-aware feedback speeds up evolving multi-part software systems
Component-Aware Feedback for Self-Evolving Programs
Artificial IntelligenceInformation Retrieval
Summary
Making complex computer programs better by randomly changing parts can be slow and confusing. The authors show a way to keep track of which part changes improve the program and by how much, so the process learns faster. They tested this on programs that sort search results to balance quality and cost, and it got better results using fewer tries. This method helps computers improve themselves more efficiently when changing many parts.
What this means in practice
- •For software engineering teams: Improve software evolution systems by providing detailed feedback on changes to multi-component programs to speed up quality improvements.
- •For search engine developers: Optimize multi-stage reranking pipelines to achieve higher ranking quality while using fewer computation resources per query.
Authors
Ethan Lin, Jinming Nian, Yi Fang
Abstract
LLM-guided evolutionary search can discover complex programs, but existing methods mostly only save candidate programs and fitness scores while discarding which component edits produced which fitness metric changes. Existing methods force the mutator LLM to infer the effect of prior edits from cluttered histories, making program search slow and unstable. This is especially true for locally servable LLMs to evolve multi-component systems. We introduce component-aware feedback, which compares each evaluated program with its parent, identifies the components that changed, and logs them with the associated metric differences into an attribution memory that later mutations read. The memory keeps each change in two reference frames, local against the parent it came from and global against the seed program, which shows both the immediate effect of a change and the cumulative progress made since the seed. We study this on LLM reranking, a multi-objective optimization problem where a multi-stage pipeline must balance quality against serving cost. Across twelve \textsc{Bright} datasets, our method reaches the strongest baseline's final quality after a median of one third of the search budget and ends 7.2\% higher in held-out nDCG@10, and under a cost-aware objective it finds pipelines that are on average more accurate while using 11\% fewer tokens per query, showing component-aware feedback to be a promising direction for more efficient self-evolving systems.