Back to the program.
Machine Learning & Optimisation: Promise and Power of Data-driven, Automated Algorithm Design
Holger H. Hoos
University of British Columbia, Vancouver, Canada
hoos@cs.ubc.ca

Abstract

Computational tools are transforming the way we live, work and interact; they also hold the key for meeting many of the challenges that we face as individuals, communities and societies. Machine learning and optimisation techniques play a particularly important role in this context, and cleverly combined, they can revolutionise the way we solve challenging computational problems - as I will demonstrate in this talk, using examples from mixed integer programming, planning and propositional satisfiability, as well as from prominent supervised machine learning tasks. The fundamental techniques I will cover include automated algorithm selection, configuration and hyperparameter optimisation, as well as performance prediction and Bayesian optimisation. I will also motivate and explain the Programming by Optimisation paradigm for automated algorithm design, which further extends the reach of those techniques.