Stochastic Models of Polymer Systems
Abstract
The stochastic gradient decent algorithm is the now the "algorithm of choice" for very large machine learning problems. We introduced the idea of "stochastic modified equation" to the analysis of such algorithms. This approach allows us to obtain very precise information about the behavior of the algorithm. At the same time, we were also able to formulate various acceleration techniques in precise math terms (e.g. formulate them as stochastic control problems) and obtain precise information about these acceleration methods. This approach is quite general and applies to other stochastic algorithms.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 2016
- Accession Number
- ADA631612
Entities
People
- Bernard Chazelle
- Cheng Tai
- Chu Wang
- Qianxiao Li
- Weinan E
Organizations
- Princeton University