File(s) under permanent embargo
Algorithmic assurance: an active approach to algorithmic testing using Bayesian optimisation
conference contribution
posted on 2018-01-01, 00:00 authored by Shivapratap Gopakumar, Sunil GuptaSunil Gupta, Santu RanaSantu Rana, Tien Vu Nguyen, Svetha VenkateshSvetha VenkateshWe introduce algorithmic assurance, the problem of testing whether machine learning algorithms are conforming to their intended design goal. We address this problem by proposing an efficient framework for algorithmic testing. To provide assurance, we need to efficiently discover scenarios where an algorithm decision deviates maximally from its intended gold standard. We mathematically formulate this task as an optimisation problem of an expensive, black-box function. We use an active learning approach based on Bayesian optimisation to solve this optimisation problem. We extend this framework to algorithms with vector-valued outputs by making appropriate modification in Bayesian optimisation via the EXP3 algorithm. We theoretically analyse our methods for convergence. Using two real-world applications, we demonstrate the efficiency of our methods. The significance of our problem formulation and initial solutions is that it will serve as the foundation in assuring humans about machines making complex decisions.
History
Event
Neural Information Processing Systems Foundation. Conference (32nd : 2018 : Montreal, Canada)Volume
31Series
Neural Information Processing Systems Foundation ConferencePagination
1 - 9Publisher
Neural Information Processing Systems FoundationLocation
Montreal, CanadaPlace of publication
San Diego, Calif.Start date
2018-12-02End date
2018-12-08ISSN
1049-5258Language
engPublication classification
E1 Full written paper - refereedCopyright notice
2018, Neural Information Processing Systems Foundation, Inc.Editor/Contributor(s)
S Bengio, H Wallach, H Larochelle, K Grauman, N CesaBianchi, R GarnettTitle of proceedings
NeurIPS 2018 : Proceedings of the 32nd Conference on Neural Information Processing SystemsUsage metrics
Categories
No categories selectedKeywords
Licence
Exports
RefWorks
BibTeX
Ref. manager
Endnote
DataCite
NLM
DC