Abstract
In this paper, a gradient-based method for bound constrained non-convex problems is proposed. By leveraging both projected gradient descent and perturbed gradient descent, the proposed algorithm, named perturbed projected gradient descent (PP-GD), converges to some approximate second-order stationary (SS2) points (which satisfy certain approximate second-order necessary conditions) with provable convergence rate guarantees. The proposed algorithm is suitable for a large-scale problem since it only uses the gradient information of the objective function. It also seamlessly incorporates variable constraints such as nonnegativity, which is commonly seen in many practical machine learning problems. We provide a concrete theoretical analysis showing that PP-GD is able to obtain approximate second-order solutions by extracting the negative curvature of the objective function around the strict saddle points. Numerical results demonstrate that PP-GD indeed converges faster compared to other first-order methods in the presence of strict saddle points.
Original language | English (US) |
---|---|
Title of host publication | 2019 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2019 - Proceedings |
Publisher | Institute of Electrical and Electronics Engineers Inc. |
Pages | 5356-5360 |
Number of pages | 5 |
ISBN (Electronic) | 9781479981311 |
DOIs | |
State | Published - May 1 2019 |
Event | 44th IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2019 - Brighton, United Kingdom Duration: May 12 2019 → May 17 2019 |
Publication series
Name | ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings |
---|---|
Volume | 2019-May |
ISSN (Print) | 1520-6149 |
Conference
Conference | 44th IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2019 |
---|---|
Country/Territory | United Kingdom |
City | Brighton |
Period | 5/12/19 → 5/17/19 |
Bibliographical note
Publisher Copyright:© 2019 IEEE.
Keywords
- Strict saddle points
- convergence rate
- projected gradient descent
- second-order stationary (SS2) points