Monday 08 Feb 2016Offline Machine Learning for Selection Hyper-Heuristics

Bill Yates -

Har/170 15:30-16:30

Selection hyper-heuristics are high level heuristic that select and apply low level heuristics. They are typically employed to solve hard search or optimisation problems, such as exam time tabling or the travelling salesman problem. Some hyper-heuristics employ machine learning techniques to further improve their performance. We can distinguish between two forms of learning; online, where learning takes place during execution and offline learning.
Offline learning takes place on a database of hyper-heuristic runs or executions over a number of problems and a number of problem domains.
In this talk we describe the construction of such a database and present the initial results of our attempts to learn or discover qualitative and quantitative information that can be used to enhance hyper-heuristic performance.

Add to calendar

Add to calendar (.ics)