뉴스 및 공지사항
- 관리자
- 2024.09.02
- 63
[세미나] [GSDS Colloquium] Sept. 3th 16:00 / Room 1501 E2 / Prof. Woochang Kim / Neural Stochastic Optimization and its Applications
GSDS 콜로퀴움이 다음과 같이 진행될 예정입니다.
# 날짜/시간: 2024년 9월 3일 수요일 16:00~17:00
# 장소: E2 1501호
# 연사: Prof. WooChang Kim (Industrial and Systems Engineering)
# 제목: Neural Stochastic Optimization and its Applications
- Zoom 링크: https://kaist.zoom.us/j/82591870965 ID: 825 9187 0965
# Abstract
Traditional approaches in solving large-scale stochastic programming problems include the use of decomposition techniques, sample average approximation, and scenario reduction methods. We propose a framework for neural stochastic optimization (NSO) that utilizes machine learning in stochastic programming. Specifically, we establish theoretical foundations for approximating value functions and optimal policies through Neural Networks by providing conditions under which the Universal Approximation Theorem can be applied to parametric optimization problems. This study addresses a gap in the analysis of when Neural Networks can be effectively used as approximators for parametric optimization and demonstrates that the approximation error decreases as the amount of training data increases. Additionally, we propose the use of transfer learning in NSO, resulting in improved performance with less data. These results suggest that NSO has the potential to provide real-time planning in a scalable manner for large scale stochastic problems. Based on this analysis, we tackle several decision making problems in energy, production and finance.
