HAL will be down for maintenance from Friday, June 10 at 4pm through Monday, June 13 at 9am. More information
Skip to Main content Skip to Navigation
Conference papers

Generalized Abs-Linear Learning by Mixed Binary Quadratic Optimization

Abstract : We consider predictor functions in generalized abs-linear form, which generalize neural nets with hinge activation. To train them with respect to a given data set of feature-label pairs, one has to minimize the average loss, which is a multi-piecewise linear or quadratic function of the weights, i.e. coefficients of the abs-linear form. We suggest to attack this nonsmooth, global optimization problem via successive piecewise linearization, which allows the application of mixed binary convex quadratic optimization codes amongst other methods. These solve the sequence of abs-linear model problems with a proximal term. Preliminary experiments on a simple regression problem verify the validity of the approach but require a large number of Simplex pivots by the solver Gurobi.
Complete list of metadata

Cited literature [9 references]  Display  Hide  Download

https://hal.archives-ouvertes.fr/hal-02945038
Contributor : Angel Adrian Rojas Jimenez Connect in order to contact the contributor
Submitted on : Monday, September 21, 2020 - 11:54:48 PM
Last modification on : Wednesday, September 22, 2021 - 2:08:01 PM
Long-term archiving on: : Thursday, December 3, 2020 - 3:32:37 PM

File

CARI2020-34.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : hal-02945038, version 1

Collections

Citation

Andreas Griewank, Angel Adrian Rojas Jimenez. Generalized Abs-Linear Learning by Mixed Binary Quadratic Optimization. Proceedings of CARI 2020, Oct 2020, Thes, Senegal. ⟨hal-02945038⟩

Share

Metrics

Record views

22

Files downloads

14