Machine learning is increasingly becoming a daily tool for natural scientists, forging collaborations across disciplines. This article presents a brief overview from a modern machine-learning viewpoint on what machine learning is, how it can be useful for natural science research, and how it can transform our way of doing science. Every natural science field is now facing diverse experimental, simulated, and literature-based data, and trying to leverage this accessibility and multiplicity of heterogenous views to full advantage. Lessons from data-centric multidisciplinary research over recent years well as common pitfalls and rising challenges are discussed.

  Deep learning becomes promising techniques in many applications such as machine translation and drug discovery. Recently, it is actively studied to use deep learning for discovering new material such as chemicals and crystals. These works contribute to efficient and effective material discovery by reducing researcher-dependent tasks and time costs. In this article, we review existing deep learning techniques on crystal discovery. We explain recent studies categorized into: classifier/regression, generation, and crystal growth. Also, we describe fundamental knowledge of deep learning for beginners: general deep learning models, data representations of crystals, and difference between drug-like materials and crystals. Furthermore, we explain databases and tools that are useful for applying deep learning on material discovery. Finally, we discuss some challenges for crystal discovery that we need to tackle in the future.

  Chemometrics, a discipline of analytical chemistry extracting information from chemical systems based on statistics and multivariate analysis, is briefly Typical data analyses in chemistry, dimensionality reduction, clustering, regression and classification, were demonstrated using multivariate data obtained by instrumental analysis such as spectra by optical method and chromatogram by separation analysis. Numerical computations in chemometrics using Python language assisted by a machine learning library of scikit-learn are also overviewed.

  This tutorial provides an overview of the research trends to date in the field of crystal growth theory of semiconductors. Furthermore, a future outlook is discussed based on the research trends to date. In the research field, the major trend to date was to analyze and understand the experimental results and phenomena theoretically. As a number of physical models have already been developed in the research field, it is expected that these models will be used to construct “digital twins” of chemical reaction system and apply them to “process informatics” to predict optimum growth condition of new materials.

  In recent years, systems and services that utilize information science and technology (so-called AI) have been widespread and are changing our lives and industries. Also in academic researches, a large number of studies have been achieved in various research fields, including the field of crystal growth. There are two types of data related to crystal growth: data generated by simulations and data generated by actual experiments. In this paper, I focus on informatics applications using data from actual crystal growth experiments. I first give an introduction of informatics applications and then discuss the difficulties in applying informatics to the crystal growth process and in using actual experimental data. Next, examples of research conducted by the authors are presented. Finally, I summarize this paper and discuss the issues and challenges of applying informatics to crystal growth research.

  In this paper, machine learning and optimization based on simulation data of crystal growth are discussed from the viewpoint of informatics application, introducing our application to solution growth of SiC crystal. First, general aspects of crystal growth process simulation and its informatics applications are described. Next, after an overview of the solution growth of SiC crystal, the process optimization of solution growth of SiC crystal using machine learning is described, including the prediction model of temperature and flow of the solution in the crucible, the optimization of geometry conditions, and the optimization of process conditions corresponding to time evolution. Next, application to other materials is described. Finally, this paper is summarized with future prospects.

  We apply a convolutional neural network (CNN) to low-dose electron images obtained by a transmission electron microscope. Using our original dataset including short-exposure images and long-exposure images, the CNN model is trained. The CNN model produced images that were equivalent to images 10000 times brighter. The particles in improved images become clearer and can be counted. The waiting time for image conversion is approximately 8 ms, and the in-situ observation is also possible.