Fast and Furious Convergence: Stochastic Second Order Methods under Interpolation
Abstract
We consider stochastic second-order methods for minimizing smooth and strongly-convex functions under an interpolation condition satisfied by over-parameterized models. Under this condition, we show that the regularized subsampled Newton method (R-SSN) achieves global linear convergence with an adaptive step-size and a constant batch-size. By growing the batch size for both the subsampled gradient and Hessian, we show that R-SSN can converge at a quadratic rate in a local neighbourhood of the solution. We also show that R-SSN attains local linear convergence for the family of self-concordant functions. Furthermore, we analyze stochastic BFGS algorithms in the interpolation setting and prove their global linear convergence. We empirically evaluate stochastic L-BFGS and a "Hessian-free" implementation of R-SSN for binary classification on synthetic, linearly-separable datasets and real datasets under a kernel mapping. Our experimental results demonstrate the fast convergence of these methods, both in terms of the number of iterations and wall-clock time.
Cite
@article{arxiv.1910.04920,
title = {Fast and Furious Convergence: Stochastic Second Order Methods under Interpolation},
author = {Si Yi Meng and Sharan Vaswani and Issam Laradji and Mark Schmidt and Simon Lacoste-Julien},
journal= {arXiv preprint arXiv:1910.04920},
year = {2020}
}
Comments
AISTATS, 2020