Reproduce “MetaDSE: A Few-shot Meta-learning Framework for Cross-workload CPU Design Space Exploration” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Runzhen Xue, Hao Wu, Mingyu Yan, Ziheng Xiao, Xiaochun Ye, Dongrui Fan
A meta-learning framework called MetaDSE enables efficient CPU design space exploration across different workloads through a combination of transformer-based prediction and workload-adaptive architectural masking, reducing prediction error by 44.3% compared to existing methods while requiring minimal adaptation data for new workloads.
Key insights
- Meta-learning enables more robust generalization across workloads compared to traditional transfer learning
- Architectural masking improves adaptation efficiency by filtering irrelevant parameter interactions
- Aligning data distribution between pre-training and adaptation phases enhances performance
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.