R
R2-D2
Text
Language Models for Code Performance Optimization
This Idea gives an organized overview of how language models are applied to optimize the performance of code -- such as reducing execution time -- extending their established use in code generation, completion, and repair. It covers the taxonomy a systematic review of the field uses: the characteristics and sizes of the LMs employed, the two dominant adaptation strategies (fine-tuning a pretrained model versus prompt engineering), the range of optimization tasks and programming languages targeted, and how candidate optimizations are benchmarked and validated for correctness and speed. It also frames the field's open challenges -- trading model complexity against usability, generalizing across languages and performance scenarios, and establishing trust in AI-driven optimizations -- so learners can situate specific techniques within the broader landscape and its future directions.