EURO-Online login
- New to EURO? Create an account
- I forgot my username and/or my password.
- Help with cookies
(important for IE8 users)
1298. Feature selection in linear SVMs: a scalable SDP decomposition approach using a hard cardinality constraint
Invited abstract in session MA-32: Large Scale Constrained Optimization: Algorithms and Applications, stream Advances in large scale nonlinear optimization.
Monday, 8:30-10:00Room: 41 (building: 303A)
Authors (first author is the speaker)
| 1. | Bo Peng
|
| University of Vienna | |
| 2. | Immanuel Bomze
|
| Dept. of Statistics and OR, University of Vienna | |
| 3. | Laura Palagi
|
| Department of Computer, Control, and Management Engineering A. Ruberti, Sapienza University of Rome |
Abstract
In this talk, we focus on the feature selection problem in linear SVMs by using a hard cardinality constraint. The problem is first reformulated into mixed-integer form, for which a novel SDP relaxation is proposed. Exploiting the sparse pattern of the relaxation, we decompose it and obtain an equivalent SDP relaxation in a three-dimensional positive semi-definite cone. Based on the decomposed SDP relaxation, we propose heuristics using the information of its optimal solution. Moreover, an exact procedure is proposed by solving a sequence of mixed-integer decomposed SDPs. Numerical results on classical online datasets are reported.
Keywords
- Large Scale Optimization
- Programming, Mixed-Integer
- Programming, Semidefinite
Status: accepted
Back to the list of papers