EURO-Online login
- New to EURO? Create an account
- I forgot my username and/or my password.
- Help with cookies
(important for IE8 users)
1797. Learning Directed Acyclic Graphs using Integer Linear Programming
Invited abstract in session MC-26: Combinatorial Optimization for Machine Learning, stream Combinatorial Optimization.
Monday, 12:30-14:00Room: 012 (building: 208)
Authors (first author is the speaker)
| 1. | James Cussens
|
| School of Computer Science, University of Bristol |
Abstract
Directed acyclic graphs (DAGs) can be used to represent conditional
independence relations in joint probability distributions, and, with
suitable assumptions, can additionally represent causal relations
between the variables in a distribution. The problem of learning the
'best' DAG from data is often cast as a discrete optimisation problem
- which is known to be NP-hard. Encoding this problem as an integer
linear program (ILP) is not hard, but solving large instances requires
a careful combination of cutting (adding linear constraints during
solving) and pricing (adding new ILP variables during solving). In
this talk I will discuss 'lessons learned' in tackling the
DAG-learning problem with ILP and discuss the more ambitious task of
learning models (causal and non-causal) which allow for latent variables.
Keywords
- Combinatorial Optimization
- Machine Learning
- Column Generation
Status: accepted
Back to the list of papers