78. Fully stochastic trust-region methods with Barzilai-Borwein steplengths
Invited abstract in session TB-3: Theoretical and algorithmic advances in large scale nonlinear optimization and applications Part 1, stream Large scale optimization: methods and algorithms.
Tuesday, 10:30-12:30Room: B100/4011
Authors (first author is the speaker)
| 1. | Benedetta Morini
|
| Dipartimento di Ingegneria Industriale, Universita di Firenze | |
| 2. | Mahsa Yousefi
|
| Department of Industrial Engineering, University of Florence | |
| 3. | Stefania Bellavia
|
| Dipartimento di Ingegneria Industriale, Universita di Firenze |
Abstract
We discuss stochastic gradient methods using stochastic adaptations of Barzilai-Borwein steplengths for finite-sum minimization problems. Our approach builds on the Trust-Region-ish (TRish) framework, a first-order stochastic trust-region method based on careful step normalization. Our framework, TRishBB, is designed to enhance the performance of TRish while reducing the computational cost of its second-order variant.
In this talk, we introduce TRishBB in three variants, each leveraging Barzilai-Borwein steplengths in a stochastic setting. We will highlight the theoretical foundations of TRishBB, key insights, and properties from the convergence analysis, and discuss its practical impact on machine learning applications with numerical results.
Keywords
- Large-scale optimization
- Stochastic optimization
- Optimization for learning and data analysis
Status: accepted
Back to the list of papers