Reproduce “A Kernel Theory of Modern Data Augmentation” 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
Tri Dao, Albert Gu, Alexander J. Ratner, Virginia Smith, Christopher De Sa, Christopher Ré
Dao et al. analyze data augmentation by averaging representations of transformed examples and adding a second-order term that accounts for their output variance. Their kernel analysis is exact under local transformations and a smooth convex loss, while the neural-network use is heuristic. In experiments, LeNet retains 92% of MNIST’s accuracy gain with 30% less estimated computation, not measured training time.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.