5 papers
A Theory of Saddle Escape in Deep Nonlinear Networks
Divit Rawal, Michael R. DeWeese
In deep networks with small initialization, training exhibits long plateaus separated by sharp feature-acquisition transitions. Whereas shallow nonlinear networks and deep linear n…
Majority-of-Three is Optimal
Divit Rawal, Nikita Zhivotovskiy
We give a short proof that the majority vote of three independent consistent classifiers is an optimal learner in the realizable PAC setting. This proves optimality for the simples…
Rao-Blackwellized Score Matching on Manifolds
Divit Rawal
We study denoising score matching (DSM) when data are drawn from an embedded manifold . We show that under ambient Gaussian corruption, the target has varia…
ALPHANSO: Open-Source Modeling of (,n) Neutron Source Terms
Divit Rawal, Anthony J. Nelson, William Zywiec +1
Applications ranging from nuclear safeguards to dark matter detection require accurate predictions of neutron yields and energy spectra produced by (,n) reactions. Legacy tools…
Minimax Rates for Hyperbolic Hierarchical Learning
Divit Rawal, Sriram Vishwanath
We prove an exponential separation in sample complexity between Euclidean and hyperbolic representations for learning on hierarchical data under standard Lipschitz regularization.…