: 13h30, ngày 29/04/2022 (Thứ Sáu)
: Machine Learning và Data Mining
: Võ Quỳnh Trang
: NCS LIMOS Université Clermont Auvergne
Tóm tắt báo cáo
Integer Linear Programming (ILP) is a versatile modeling method for combinatorial optimization problems. Since ILP is hard to solve in theory(NP-hard) and practice, modern ILP solvers rely on heuristics. Machine Learning (ML) arises as a natural tool for tunning these heuristics. This seminar introduces several machine-learning-based approaches to improve the performance of ILP solvers. In the First part, we present the branch-and-bound framework for ILP. In the second part, we talk about machine learning applications
for controlling problems during the solving process. Finally, we introduce Reinforcement Learning perspectives to learn solver search policies.