Nonlinear stochastic optimization plays a vital role in a wide range of applications, includ- ing machine learning, statistics, energy, and signal processing. In machine learning, models are formulated parametrically and those parameters are optimized to best predict output for yet unseen data. Some machine learning problems are distinguished by the immense scale of data they face and by the high dimensional nature of the parameters requiring optimization. The goal of this proposal is to develop new algorithms for large-scale stochastic optimization prob- lems of this kind. The proposal describes a broad research agenda of creating new algorithms that are more e�cient and more robust than classical methods, and are better able to address the challenges unique to data science applications. The proposal consists of three projects, two of which are thematically related as they focus on stochastic problems, and a project that addresses non-smooth structured deterministic optimization problems. In all cases, the goal is to design and implement algorithms in software, and test them on realistic applications.
|Effective start/end date||8/1/16 → 7/31/20|
- National Science Foundation (DMS-1620022)
Explore the research topics touched on by this project. These labels are generated based on the underlying awards/grants. Together they form a unique fingerprint.