Learning physical properties of liquid crystals with deep convolutional neural networks

Higor Y D Sigaki,Ervin K Lenzi,Matjaž Perc,Haroldo V Ribeiro,Rafael S Zola

doi:10.1038/s41598-020-63662-9

Abstract

Machine learning algorithms have been available since the 1990s, but it is much more recently that they have come into use also in the physical sciences. While these algorithms have already proven to be useful in uncovering new properties of materials and in simplifying experimental protocols, their usage in liquid crystals research is still limited. This is surprising because optical imaging techniques are often applied in this line of research, and it is precisely with images that machine learning algorithms have achieved major breakthroughs in recent years. Here we use convolutional neural networks to probe several properties of liquid crystals directly from their optical images and without using manual feature engineering. By optimizing simple architectures, we find that convolutional neural networks can predict physical properties of liquid crystals with exceptional accuracy. We show that these deep neural networks identify liquid crystal phases and predict the order parameter of simulated nematic liquid crystals almost perfectly. We also show that convolutional neural networks identify the pitch length of simulated samples of cholesteric liquid crystals and the sample temperature of an experimental liquid crystal with very high precision.

Highlights

Machine learning algorithms have been available since the 1990s, but it is much more recently that they have come into use in the physical sciences
The development and the spread of these machine learning methods in combination with the increasing availability of large data sets has been recognized as the “fourth paradigm of science”[40] and the “fourth industrial revolution”[41], having great potential to significantly enhance the role of computational methods in applied and fundamental research
We show that these convolutional networks classify the pitch length of simulated cholesteric liquid crystals with virtually 100% accuracy, and we identify the sample temperature of an experimental liquid crystal (E7 mixture) with impressive precision

Summary

Introduction

Machine learning algorithms have been available since the 1990s, but it is much more recently that they have come into use in the physical sciences. The introduction of a deep neural network model (AlexNet) by Krizhevsky et al.[25] in 2012 is considered the major breakthrough in the competition because the top-5 error rate was www.nature.com/scientificreports reduced from 26% to 16.4%, but mainly due to the fact that deep learning algorithms became the top contestants ever since[24] It was a deep learning approach (ResNet) that surpassed human-level performance in the ImageNet data set for the first time in 201526,27. The development and the spread of these machine learning methods in combination with the increasing availability of large data sets has been recognized as the “fourth paradigm of science”[40] and the “fourth industrial revolution”[41], having great potential to significantly enhance the role of computational methods in applied and fundamental research Despite this unparalleled surge of applications of machine learning methods in the physical sciences, there are still several areas that have taken little to no advantage of these approaches. This is somewhat surprising because, like many biological and other complex materials, the physical properties of liquid crystals are often probed by means of imaging techniques such as polarized optical microscope imaging[43]

Objectives

Methods

Results

Conclusion