Reproduce “Deep Learning Theory through the Lens of Statistical Physics: A Topical Review” 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

Wei Huang, Youjin Deng

The authors review statistical-physics accounts of how deep networks learn, focusing on loss landscapes, wide-network limits, and stochastic training dynamics. They contrast fixed-kernel Neural Tangent Kernel training, where features change little, with mean-field and dynamical mean-field theories that track evolving features. The review organizes complementary theories and their assumptions; their crossover to realistic finite-width training remains unresolved.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.