Use this URL to cite or link to this record in EThOS:
Title: Efficient pairwise information collection for subset selection
Author: Groves, Matthew J.
ISNI:       0000 0004 9358 3507
Awarding Body: University of Warwick
Current Institution: University of Warwick
Date of Award: 2020
Availability of Full Text:
Access from EThOS:
Access from Institution:
In this work, we consider the problems of selecting the subset of the top-k best of a set of alternatives, where the fitness of alternatives must be estimated through noisy pairwise sampling. To do this, we propose two novel active pairwise sampling methods, adapted from popular non-pairwise ranking and selection frameworks. We prove that our proposed methods have desirable asymptotic properties, and demonstrate empirically that they can perform better than current state-of-the art pairwise selection algorithms on a range of tasks. We show how our proposed methods can be integrated into the Covariance Matrix Adaptation Evolutionary Strategy, to improve fitness evaluation and optimizer performance including in the evolution of neural network based agents for playing No Limit Texas Hold’em poker. Finally, we demonstrate how parametric models can be used to help our proposed sampling algorithms exploit transitive preference structure between alternative pairs.
Supervisor: Not available Sponsor: Engineering and Physical Sciences Research Council ; Association for Computing Machinery
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available
Keywords: QA Mathematics