Astro for Derivative-Based Stochastic Optimization: Algorithm Description & Numerical Experiments
Abstrak
Adaptive Sampling Trust-Region Optimization (ASTRO) is a class of derivative-based stochastic trust-region algorithms developed to solve stochastic unconstrained optimization problems where the objective function and its gradient are observable only through a noisy oracle or using a large dataset. ASTRO incorporates adaptively sampled function and gradient estimates within a trust-region framework to generate iterates that are guaranteed to converge almost surely to a first-order or a second-order critical point of the objective function. Efficiency in ASTRO stems from two key aspects: (i) adaptive sampling to ensure that the objective function and its gradient are sampled only to the extent needed, so that small sample sizes result when iterates are far from a critical point and large sample sizes result when iterates are near a critical point; and (ii) quasi-Newton Hessian updates using BFGS. We describe ASTRO in detail, give a sense of its theoretical guarantees, and report extensive numerical results.
Topik & Kata Kunci
Penulis (3)
Daniel Vasquez
R. Pasupathy
S. Shashaani
Akses Cepat
PDF tidak tersedia langsung
Cek di sumber asli →- Tahun Terbit
- 2019
- Bahasa
- en
- Total Sitasi
- 4×
- Sumber Database
- Semantic Scholar
- DOI
- 10.1109/WSC40007.2019.9004904
- Akses
- Open Access ✓