Relating Training Instances to Automatic Design of Algorithms for Bin Packing via Features

Citation

Brownlee A, Woodward JR & Veerapen N (2018) Relating Training Instances to Automatic Design of Algorithms for Bin Packing via Features. In: Proceedings of GECCO 2018. Genetic and Evolutionary Computation Conference 2018, 15.07.2018-19.07.2018. New York: ACM, pp. 135-136. https://doi.org/10.1145/3205651.3205748

Abstract
Automatic Design of Algorithms (ADA) treats algorithm choice and design as a machine learning problem, with problem instances as training data. However, this paper reveals that, as with classification and regression, for ADA not all training sets are equally valuable. We apply genetic programming ADA for bin packing to sev- eral new and existing benchmark sets. Using sets with narrowly- distributed features for training results in highly specialised al- gorithms, whereas those with well-spread features result in very general algorithms. Variance in certain features has a strong corre- lation with the generality of the trained policies.

Keywords
Automatic design of algorithms; features; bin packing

• relating-training-instances-forrepository.pdf