Reproduce “Application of discrete Ricci curvature in pruning randomly wired neural networks: A case study with chest x-ray classification of COVID-19” 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
Pavithra Elumalai, Sudharsan Vijayaraghavan, Madhumita Mondal, Areejit Samal
A study from The Institute of Mathematical Sciences applies discrete Ricci curvature measures to prune Randomly Wired Neural Networks, achieving up to 4.161x compression and 4.472x theoretical speedup without compromising COVID-19 chest X-ray classification performance. The research highlights Forman-Ricci curvature as a computationally efficient alternative to Ollivier-Ricci curvature.
Key insights
- Discrete Ricci curvature measures provide a principled, geometry-inspired approach for neural network pruning, effectively leveraging the inherent graph structure.
- Forman-Ricci curvature (FRC) can achieve pruning effectiveness comparable to Ollivier-Ricci curvature (ORC) with significantly higher computational efficiency, particularly in ER and WS networks.
- Pruning consistently alters network topology by increasing modularity and decreasing global efficiency, indicating a trade-off between structural segregation and efficient global information flow.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.