File(s) under permanent embargo
Cascade Bayesian optimization
conference contribution
posted on 2016-12-08, 00:00 authored by Thanh Dai Nguyen, Sunil GuptaSunil Gupta, Santu RanaSantu Rana, Tien Vu Nguyen, Svetha VenkateshSvetha Venkatesh, K J Deane, P G SandersMulti-stage cascade processes are fairly common, especially in manufacturing industry. Precursors or raw materials are transformed at each stage before being used as the input to the next stage. Setting the right control parameters at each stage is important to achieve high quality products at low cost. Finding the right parameters via trial and error approach can be time consuming. Bayesian optimization is an efficient way to optimize costly black-box function. We extend the standard Bayesian optimization approach to the cascade process through formulating a series of optimization problems that are solved sequentially from the final stage to the first stage. Epistemic uncertainties are effectively utilized in the formulation. Further, cost of the parameters are also included to find cost-efficient solutions. Experiments performed on a simulated testbed of Al-Sc heat treatment through a three-stage process showed considerable efficiency gain over a naïve optimization approach.
History
Event
Advances in Artificial Intelligence. Australasian Joint Conference (29th : 2016 : Hobart, Tasmania)Volume
9992Series
Lecture notes in artificial intelligencePagination
268 - 280Publisher
Springer International PublishingLocation
Hobart, TasmaniaPlace of publication
Cham, SwitzerlandPublisher DOI
Start date
2016-12-05End date
2016-12-08ISSN
0302-9743eISSN
1611-3349ISBN-13
9783319501260Language
engPublication classification
E Conference publication; E1 Full written paper - refereedCopyright notice
2016, Springer InternationalEditor/Contributor(s)
B Kang, Q BaiTitle of proceedings
AI 2016: Advances in Artificial Intelligence : Proceedings of the Australasian Joint ConferenceUsage metrics
Categories
No categories selectedKeywords
Licence
Exports
RefWorks
BibTeX
Ref. manager
Endnote
DataCite
NLM
DC