What is it about?
Compilers often optimize the programs they produce using a common set of hand-tuned heuristics - can machines do better? Recent work has gone into applying reinforcement learning to guide how compilers optimize programs during the compilation process, and this work explores using neuroevolution as an alternative method and finds it competitive with other techniques.
Featured Image
Photo by Sangharsh Lohakare on Unsplash
Why is it important?
This work shows that neuroevolution is capable of optimizing programs based on their characteristics to be more space efficient than other reinforcement learning techniques or the default heuristics used in the LLVM compiler.
Perspectives
This was a fun line of research to explore and I would encourage others working at the intersection of compilers and reinforcement learning to consider evolutionary approaches as well.
Kade Heckel
University of Sussex
Read the Original
This page is a summary of: Neuroevolutionary Compiler Control for Code Optimization, July 2023, ACM (Association for Computing Machinery),
DOI: 10.1145/3583133.3596380.
You can read the full text:
Contributors
The following have contributed to this page
