Energy landscapes for machine learning

Andrew J. Ballard, Ritankar Das, Stefano Martiniani, Dhagash Mehta, Levent Sagun, Jacob D. Stevenson, David J. Wales

Research output: Contribution to journalReview articlepeer-review

36 Scopus citations

Abstract

Machine learning techniques are being increasingly used as flexible non-linear fitting and prediction tools in the physical sciences. Fitting functions that exhibit multiple solutions as local minima can be analysed in terms of the corresponding machine learning landscape. Methods to explore and visualise molecular potential energy landscapes can be applied to these machine learning landscapes to gain new insight into the solution space involved in training and the nature of the corresponding predictions. In particular, we can define quantities analogous to molecular structure, thermodynamics, and kinetics, and relate these emergent properties to the structure of the underlying landscape. This Perspective aims to describe these analogies with examples from recent applications, and suggest avenues for new interdisciplinary research.

Original languageEnglish (US)
Pages (from-to)12585-12603
Number of pages19
JournalPhysical Chemistry Chemical Physics
Volume19
Issue number20
DOIs
StatePublished - 2017
Externally publishedYes

Bibliographical note

Funding Information:
It is a pleasure to acknowledge discussions with Prof. Daan Frenkel, Dr Victor Ruehle, Dr Peter Wirnsberger, Prof. G?rard Ben Arous, and Prof. Yann Lecun. This research was funded by EPSRC grant EP/I001352/1, the Gates Cambridge Trust, and the ERC. DM was in the Department of Applied and Computational Mathematics and Statistics when this work was performed, and his current affiliation is Department of Systems, United Technologies Research Center, East Hartford, CT, USA.

Publisher Copyright:
© 2017 the Owner Societies.

Fingerprint Dive into the research topics of 'Energy landscapes for machine learning'. Together they form a unique fingerprint.

Cite this