Data science has emerged as the fourth scientific method, following experiment, theory, and computation. Since the launch of the Material Genome Initiative in 2011, data-driven approaches such as materials informatics and process informatics have been actively applied in materials. To integrate data science into traditional research, it is essential to understand research activities as a data cycle consisting of three phases : data generation, accumulation, and utilization. This cyclic process enhances the efficiency and scope of scientific research. To illustrate the impact of data science in materials research, this paper introduces Bayesian optimization for autonomous experimental system, machine learning potentials, personal databases using JSON format, and high-throughput automatic spectral analysis. These approaches contribute to the advancement of materials science through data-driven methodologies, accelerating the data cycle.

In recent years, “Materials Informatics,” which leverages data science for the exploration of novel materials, has gained significant attention. Among various approaches, “autonomous materials discovery,” which integrates robotics and materials simulations with machine learning, has emerged as a promising method for efficiently navigating vast material spaces. In this study, we employed a simulation-based autonomous materials discovery approach that combines first-principles calculations with Bayesian optimization to explore a broad space of magnetic alloys. As a result, we successfully discovered and synthesized novel alloys with high magnetization.

Data analysis based on machine learning using a set of spectroscopically obtained data was reviewed. In general, the number of spectral variables, such as discretized wavelength and wavenumber, tends to be large compared to the number of obtained spectra. Therefore, in order to reduce the number of explanatory variables to be smaller than the number of samples, dimensionality reduction is used to compress the spectral dataset. A practical application of dimensionality reduction by principal component analysis (PCA) using a set of attenuated total reflection infrared (ATR-IR) spectra of Japanese paper is introduced. Also, practical demonstrations of spectroscopic regression by partial least squares (PLS) and classification by support vector machine (SVM) using a set of near-infrared spectra of grapes, which hardly show remarkable peaks for the quantitative determination and qualitative classification, are reported.

The field of materials exploration is rapidly expanding to meet the social demand for functional materials. Autonomous materials research, which employs artificial intelligence (AI)-based decision-making in combination with automated synthesis and measurements carried out by robots, presents a promising avenue. Recently, significant progress has been made in the development of solid-state systems, in parallel with liquid systems where materials are more easily handled. Here, I present our recent research on the autonomous synthesis of functional inorganic oxide thin films. Through iterative operations of automated thin film deposition (utilizing robots), measurement of electrical conductivity (also performed by robots), and the application of Bayesian optimization (AI) for decision-making, we achieved approximately a tenfold increase in throughput. Furthermore, I will present trends in other countries and discuss future prospects.

Microscopic image plays a crucial information in functional design, particularly in data-driven materials science. However, extracting meaningful insights from imaging data remains a challenge. This study proposes the Extended Landau Free Energy Model, which integrates persistent homology and machine learning to quantitatively analyze magnetization reversal and pinning mechanisms in nanoscale materials. Using persistent homology, we extract topological features from domain structures and apply principal component analysis (PCA) to construct an energy landscape that describes magnetization reversal dynamics. Ridge regression is employed to correlate extracted features with magnetostatic and exchange energies, enabling a decomposition of energy barriers into their respective interactions. Our results clarify the pinning mechanism by quantifying the contributions of static and exchange interactions, and visualizing their spatial distribution via Hadamard product-based analysis. This method provides a detailed representation of pinning and depinning processes, revealing how structural heterogeneities influence energy barriers. Beyond magnetism, this framework offers broad applications in nanotechnology and data-driven materials science.

This study introduces the application of neural network potentials (NNPs) in materials science, with a focus on their use in analyzing partial crystallization and ion conduction mechanisms in glass structures, while also highlighting recent advancements in NNPs. Specifically, molecular dynamics (MD) simulations employing NNPs were performed to investigate the crystallization process of Li₃PS₄ glass under heat treatment. The simulations revealed the nucleation and growth of crystalline phases within the glass matrix, providing atomic-level insights into the crystallization mechanism. Furthermore, an assessment of the impact of crystallization on lithium transport properties demonstrated that the precipitated crystalline phase corresponded to the high-temperature α-phase. Additionally, the formation of new ion conduction pathways through the interconnection of crystal nuclei was identified as a key factor in enhancing ionic conductivity in glass-ceramic materials.

The conductance at the junction of a pipe with a different radius pipe in the vacuum piping system is calculated by Monte Carlo simulation. The resulting conductance C_ab is the same whether the net flow is from the small pipe of radius a to the large pipe of radius b or vice versa and is shown as C_ab＝2C_0a/(1－(C_0a/C_0b)), where C_0a and C_0b are orifice conductance of radius a and b, respectively. This equation indicates that C_ab＝2C_0a in the case of b＝∞ i.e. C_0b＝∞. These results are consistent of the previous conclusion that there are twice the conductance of the orifice conductance on both ends of the pipe [H. Yamakawa : Vac. Surf. Sci. 65, 478 (2022)]. The conductance C_oab at the junction of an orifice of radius a with a large pipe of radius b is also calculated and obtained as C_oab＝2C_0a/(2－(C_0a/C_0b)).