SONIA: A Symmetric Blockwise Truncated Optimization Algorithm
Jahani, Majid, Nazari, Mohammadreza, Tappenden, Rachael, Berahas, Albert S., Takáč, Martin
This work presents a new algorithm for empirical risk minimization. The algorithm bridges the gap between first- and second-order methods by computing a search direction that uses a second-order-type update in one subspace, coupled with a scaled steepest descent step in the orthogonal complement. To this end, partial curvature information is incorporated to help with ill-conditioning, while simultaneously allowing the algorithm to scale to the large problem dimensions often encountered in machine learning applications. Theoretical results are presented to confirm that the algorithm converges to a stationary point in both the strongly convex and nonconvex cases. A stochastic variant of the algorithm is also presented, along with corresponding theoretical guarantees. Numerical results confirm the strengths of the new approach on standard machine learning problems.
Jun-6-2020
- Country:
- North America > United States (0.67)
- Genre:
- Research Report > New Finding (0.50)
- Industry:
- Education (0.48)
- Technology: