See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/332010806

Machine learning and the physical sciences

Preprint · March 2019

DOI: 10.48550/arXiv.1903.10563

CITATION READS
1 4,113

8 authors, including:

Giuseppe Carleo Naftali Tishby

Swiss Federal Institute of Technology in Lausanne Hebrew University of Jerusalem
158 PUBLICATIONS 11,813 CITATIONS 290 PUBLICATIONS 21,277 CITATIONS

SEE PROFILE SEE PROFILE

Leslie Vogt
New York University
52 PUBLICATIONS 8,553 CITATIONS

SEE PROFILE

All content following this page was uploaded by Naftali Tishby on 30 November 2019.

The user has requested enhancement of the downloaded file.

 Machine learning and the physical sciences
Giuseppe Carleo
Center for Computational Quantum Physics, Flatiron Institute,
162 5th Avenue, New York, NY 10010, USA∗

Ignacio Cirac
Max-Planck-Institut fur Quantenoptik,
Hans-Kopfermann-Straße 1, D-85748 Garching, Germany

Kyle Cranmer
arXiv:1903.10563v1 [physics.comp-ph] 25 Mar 2019

Center for Cosmology and Particle Physics, Center of Data Science,

New York University, 726 Broadway, New York, NY 10003, USA

Laurent Daudet
LightOn, 2 rue de la Bourse, F-75002 Paris, France

Maria Schuld
University of KwaZulu-Natal, Durban 4000, South Africa
National Institute for Theoretical Physics, KwaZulu-Natal, Durban 4000, South Africa,
and Xanadu Quantum Computing, 777 Bay Street, M5B 2H7 Toronto, Canada

Naftali Tishby
The Hebrew University of Jerusalem, Edmond Safra Campus, Jerusalem 91904, Israel

Leslie Vogt-Maranto
Department of Chemistry, New York University, New York, NY 10003, USA

Lenka Zdeborová
Institut de physique théorique, Université Paris Saclay, CNRS, CEA,
F-91191 Gif-sur-Yvette, France†

Machine learning encompasses a broad range of algorithms and modeling tools used for a vast array
of data processing tasks, which has entered most scientific disciplines in recent years. We review in
a selective way the recent research on the interface between machine learning and physical sciences.
This includes conceptual developments in machine learning (ML) motivated by physical insights,
applications of machine learning techniques to several domains in physics, and cross-fertilization
between the two fields. After giving basic notion of machine learning methods and principles, we
describe examples of how statistical physics is used to understand methods in ML. We then move
to describe applications of ML methods in particle physics and cosmology, quantum many body
physics, quantum computing, and chemical and material physics. We also highlight research and
development into novel computing architectures aimed at accelerating ML. In each of the sections
we describe recent successes as well as domain-specific methodology and challenges.

Contents

I. Introduction 2
A. Concepts in machine learning 3
1. Supervised learning and neural networks 4
2. Unsupervised learning and generative modelling 5
3. Reinforcement learning 6

II. Statistical Physics 6

A. Historical note 6
B. Theoretical puzzles in deep learning 7
C. Statistical physics of unsupervised learning 7
1. Contributions to understanding basic unsupervised methods 7
2. Restricted Boltzmann machines 8
3. Modern unsupervised and generative modelling 9
D. Statistical physics of supervised learning 9
1. Perceptron and GLMs 9
2. Physics results on multi-layer neural networks 10
 2

3. Information Bottleneck 10
4. Landscapes and glassiness of deep learning 11
E. Applications of ML in Statistical Physics 11
F. Outlook and Challenges 12

III. Particle Physics and Cosmology 12

A. The role of the simulation 12
B. Classification and regression in particle physics 13
1. Jet Physics 14
2. Neutrino physics 14
3. Robustness to systematic uncertainties 15
4. Triggering 15
5. Theoretical particle physics 16
C. Classification and regression in cosmology 16
1. Photometric Redshift 16
2. Gravitational lens finding and parameter estimation 17
3. Other examples 17
D. Inverse Problems and Likelihood-free inference 18
1. Likelihood-free Inference 19
2. Examples in particle physics 19
3. Examples in Cosmology 20
E. Generative Models 20
F. Outlook and Challenges 21

IV. Many-Body Quantum Matter 22

A. Neural-Network quantum states 22
1. Representation theory 23
2. Learning from data 23
3. Variational Learning 24
B. Speed up many-body simulations 24
C. Classifying many-body quantum phases 25
1. Synthetic data 25
2. Experimental data 26
D. Tensor networks for machine learning 26
E. Outlook and Challenges 27

V. Quantum computing 27
A. Quantum state tomography 28
B. Controlling and preparing qubits 29
C. Error correction 29

VI. Chemistry and Materials 30

A. Energies and forces based on atomic environments 30
B. Potential and free energy surfaces 31
C. Materials properties 31
D. Electron densities for density functional theory 32
E. Data set generation 32
F. Outlook and Challenges 34

VII. AI acceleration with classical and quantum hardware 34

A. Beyond von Neumann architectures 34
B. Neural networks running on light 34
C. Revealing features in data 35
D. Quantum-enhanced machine learning 35
E. Outlook and Challenges 36

VIII. Conclusions and Outlook 36

Acknowledgements 36

References 37

I. INTRODUCTION

∗ Corresponding author: [email protected] The past decade has seen a prodigious rise of machine-
† Corresponding author: [email protected] learning (ML) based techniques, impacting many areas
in industry including autonomous driving, health-care, fi-
nance, manufacturing, energy harvesting, and more. ML
 3

is largely perceived as one of the main disruptive tech- Section III treats progress in the fields of high-energy
nologies of our ages, as much as computers have been physics and cosmology, Section IV reviews how ML ideas
in the 1980’s and 1990’s. The general goal of ML is to are helping to understand the mysteries of many-body
recognize patterns in data, which inform the way unseen quantum systems, Section V briefly explore the promises
problems are treated. For example, in a highly complex of machine learning within quantum computations, and
system such as a self-driving car, vast amounts of data in Section VI we highlight some of the amazing advances
coming from sensors have to be turned into decisions of in computational chemistry and materials design due to
how to control the car by a computer that has “learned” ML applications. In Section VII we discuss some ad-
to recognize the pattern of “danger”. vances in instrumentation leading potentially to hard-
The success of ML in recent times has been marked ware adapted to perform machine learning tasks. We
at first by significant improvements on some existing conclude with an outlook in Section VIII.
technologies, for example in the field of image recogni-
tion. To a large extent, these advances constituted the
first demonstrations of the impact that ML methods can
have in specialized tasks. More recently, applications tra-
ditionally inaccessible to automated software have been
successfully enabled, in particular by deep learning tech- A. Concepts in machine learning
nology. The demonstration of reinforcement learning
techniques in game playing, for example, has had a deep For the purpose of this review we will briefly explain
impact in the perception that the whole field was moving some fundamental terms and concepts used in machine
a step closer to what expected from a general artificial learning. For further reading, we recommend a few re-
intelligence. sources, some of which have been targeted especially for
In parallel to the rise of ML techniques in industrial ap- a physics audience. For a historical overview of the
plications, scientists have increasingly become interested development of the field we recommend Refs. (LeCun
in the potential of ML for fundamental research, and et al., 2015; Schmidhuber, 2014). An excellent recent
physics is no exception. To some extent, this is not too introduction to machine learning for physicists is Ref.
surprising, since both ML and physics share some of their (Mehta et al., 2018), which includes notebooks with prac-
methods as well as goals. The two disciplines are both tical demonstrations. A very useful online resource is
concerned about the process of gathering and analyzing Florian Marquardt’s course “Machine learning for physi-
data to design models that can predict the behaviour of cists” 1 . Useful textbooks written by machine learning
complex systems. However, the fields prominently differ researchers are Christopher Bishop’s standard textbook
in the way their fundamental goals are realized. On the (Bishop, 2006), as well as (Goodfellow et al., 2016) which
one hand, physicists want to understand the mechanisms focuses on the theory and foundations of deep learning
of Nature, and are proud of using their own knowledge, and covers many aspects of current-day research. A vari-
intelligence and intuition to inform their models. On ety of online tutorials and lectures is useful to get a basic
the other hand, machine learning mostly does the oppo- overview and get started on the topic.
site: models are agnostic and the machine provides the
’intelligence’ by extracting it from data. Although of- To learn about the theoretical progress made in statis-
ten powerful, the resulting models are notoriously known tical physics of neural networks in the 1980s-1990s we rec-
to be as opaque to our understanding as the data pat- ommend the rather accessible book Statistical Mechan-
terns themselves. Machine learning tools in physics are ics of Learning (Engel and Van den Broeck, 2001). For
therefore welcomed enthusiastically by some, while being learning details of the replica method and its use in com-
eyed with suspicions by others. What is difficult to deny puter science, information theory and machine learning
is that they produce surprisingly good results in some we would recommend the book of Nishimori (Nishimori,
cases. 2001). For the more recent statistical physics methodol-
ogy the textbook of Mézard and Montanari is an excellent
In this review, we attempt at providing a coherent se-
reference (Mézard and Montanari, 2009).
lected account of the diverse intersections of ML with
physics. Specifically, we look at an ample spectrum of To get a basic idea of the type of problems that ma-
fields (ranging from statistical and quantum physics to chine learning is able to tackle it is useful to defined three
high energy and cosmology) where ML recently made a large classes of learning problems: Supervised learning,
prominent appearance, and discuss potential applications unsupervised learning and reinforcement learning. This
and challenges of ‘intelligent’ data mining techniques in will also allow us to state the basic terminology, build-
the different contexts. We start this review with the field ing basic equipment to expose some of the basic tools of
of statistical physics in Section II where the interaction machine learning.
with machine learning has a long history, drawing on
methods in physics to provide better understanding of
problems in machine learning. We then turn the wheel in
the other direction of using machine learning for physics. 1 See https://machine-learning-for-physicists.org/.