Projective Preferential Bayesian Optimization

Petrus Mikkola; Milica Todorovic; Jari Järvi; Patrick Rinke; Samuel Kaski

Projective Preferential Bayesian Optimization

Petrus Mikkola, Milica Todorovic, Jari Järvi, Patrick Rinke, Samuel Kaski

Research output: Chapter in Book/Report/Conference proceeding › Conference article in proceedings › Scientific › peer-review

6 Citations (Scopus)

85 Downloads (Pure)

Abstract

Bayesian optimization is an effective method for finding extrema of a black-box function. We propose a new type of Bayesian optimization for learning user preferences in high-dimensional spaces. The central assumption is that the underlying objective function cannot be evaluated directly, but instead a minimizer along a projection can be queried, which we call a projective preferential query. The form of the query allows for feedback that is natural for a human to give, and which enables interaction. This is demonstrated in a user experiment in which the user feedback comes in the form of optimal position and orientation of a molecule adsorbing to a surface. We demonstrate that our framework is able to find a global minimum of a high-dimensional black-box function, which is an infeasible task for existing preferential Bayesian optimization frameworks that are based on pairwise comparisons.

Original language	English
Title of host publication	37th International Conference on Machine Learning, ICML 2020
Publisher	International Machine Learning Society
Pages	6840-6848
Number of pages	9
ISBN (Electronic)	9781713821120
Publication status	Published - 21 Nov 2020
MoE publication type	A4 Conference publication
Event	International Conference on Machine Learning - Vienna, Austria Duration: 12 Jul 2020 → 18 Jul 2020 Conference number: 37

Publication series

Name	Proceedings of Machine Learning Research
ISSN (Electronic)	2640-3498

Conference

Conference	International Conference on Machine Learning
Abbreviated title	ICML
Country/Territory	Austria
City	Vienna
Period	12/07/2020 → 18/07/2020

Keywords

human-in-the-loop machine learning
gaussian process
preference learning
Bayesian optimization
Bayesian methods
machine learning
expert elicitation

Access to Document

Mikkola_Projective Preferential Bayesian Optimization.20a-2Final published version, 1.44 MB

http://proceedings.mlr.press/v119/mikkola20a.html

Cite this

@inproceedings{809721efe2fe40e3b0a19b105dea82be,

title = "Projective Preferential Bayesian Optimization",

abstract = "Bayesian optimization is an effective method for finding extrema of a black-box function. We propose a new type of Bayesian optimization for learning user preferences in high-dimensional spaces. The central assumption is that the underlying objective function cannot be evaluated directly, but instead a minimizer along a projection can be queried, which we call a projective preferential query. The form of the query allows for feedback that is natural for a human to give, and which enables interaction. This is demonstrated in a user experiment in which the user feedback comes in the form of optimal position and orientation of a molecule adsorbing to a surface. We demonstrate that our framework is able to find a global minimum of a high-dimensional black-box function, which is an infeasible task for existing preferential Bayesian optimization frameworks that are based on pairwise comparisons.",

keywords = "human-in-the-loop machine learning, gaussian process, preference learning, Bayesian optimization, Bayesian methods, machine learning, expert elicitation",

author = "Petrus Mikkola and Milica Todorovic and Jari J{\"a}rvi and Patrick Rinke and Samuel Kaski",

year = "2020",

month = nov,

day = "21",

language = "English",

series = "Proceedings of Machine Learning Research",

publisher = "International Machine Learning Society ",

pages = "6840--6848",

booktitle = "37th International Conference on Machine Learning, ICML 2020",

address = "United States",

note = "International Conference on Machine Learning, ICML ; Conference date: 12-07-2020 Through 18-07-2020",

}

Mikkola, P, Todorovic, M, Järvi, J, Rinke, P & Kaski, S 2020, Projective Preferential Bayesian Optimization. in 37th International Conference on Machine Learning, ICML 2020. Proceedings of Machine Learning Research, International Machine Learning Society , pp. 6840-6848, International Conference on Machine Learning, Vienna, Austria, 12/07/2020. <http://proceedings.mlr.press/v119/mikkola20a.html>

Projective Preferential Bayesian Optimization. / Mikkola, Petrus; Todorovic, Milica; Järvi, Jari et al.
37th International Conference on Machine Learning, ICML 2020. International Machine Learning Society , 2020. p. 6840-6848 (Proceedings of Machine Learning Research).

Research output: Chapter in Book/Report/Conference proceeding › Conference article in proceedings › Scientific › peer-review

TY - GEN

T1 - Projective Preferential Bayesian Optimization

AU - Mikkola, Petrus

AU - Todorovic, Milica

AU - Järvi, Jari

AU - Rinke, Patrick

AU - Kaski, Samuel

N1 - Conference code: 37

PY - 2020/11/21

Y1 - 2020/11/21

N2 - Bayesian optimization is an effective method for finding extrema of a black-box function. We propose a new type of Bayesian optimization for learning user preferences in high-dimensional spaces. The central assumption is that the underlying objective function cannot be evaluated directly, but instead a minimizer along a projection can be queried, which we call a projective preferential query. The form of the query allows for feedback that is natural for a human to give, and which enables interaction. This is demonstrated in a user experiment in which the user feedback comes in the form of optimal position and orientation of a molecule adsorbing to a surface. We demonstrate that our framework is able to find a global minimum of a high-dimensional black-box function, which is an infeasible task for existing preferential Bayesian optimization frameworks that are based on pairwise comparisons.

AB - Bayesian optimization is an effective method for finding extrema of a black-box function. We propose a new type of Bayesian optimization for learning user preferences in high-dimensional spaces. The central assumption is that the underlying objective function cannot be evaluated directly, but instead a minimizer along a projection can be queried, which we call a projective preferential query. The form of the query allows for feedback that is natural for a human to give, and which enables interaction. This is demonstrated in a user experiment in which the user feedback comes in the form of optimal position and orientation of a molecule adsorbing to a surface. We demonstrate that our framework is able to find a global minimum of a high-dimensional black-box function, which is an infeasible task for existing preferential Bayesian optimization frameworks that are based on pairwise comparisons.

KW - human-in-the-loop machine learning

KW - gaussian process

KW - preference learning

KW - Bayesian optimization

KW - Bayesian methods

KW - machine learning

KW - expert elicitation

UR - https://arxiv.org/abs/2002.03113

M3 - Conference article in proceedings

T3 - Proceedings of Machine Learning Research

SP - 6840

EP - 6848

BT - 37th International Conference on Machine Learning, ICML 2020

PB - International Machine Learning Society

T2 - International Conference on Machine Learning

Y2 - 12 July 2020 through 18 July 2020

ER -

Projective Preferential Bayesian Optimization

Abstract

Publication series

Conference

Keywords

Access to Document

Other files and links

Fingerprint

Interactive machine learning from multiple biodata sources

-: Finnish Center for Artificial Intelligence

White-boxed artificial intelligence

Science-IT

Cite this

Projective Preferential Bayesian Optimization

Abstract

Publication series

Conference

Keywords

Access to Document

Other files and links

Fingerprint

Projects

Interactive machine learning from multiple biodata sources

-: Finnish Center for Artificial Intelligence

White-boxed artificial intelligence

Equipment

Science-IT

Cite this