HKHiromasa Kaneko

Hiromasa Kaneko

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Hiromasa Kaneko has a total of 159 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Applicability-domain-boundary-aware molecular generation for property/activity optimization

    by Yuta Uchibori, Tatsuhito Ando, Nobuyuki N. Matsuzawa, Hiroyuki Maeshima, Hiromasa Kaneko

    In the molecular design of high-performance materials, physical properties or activities (y) of chemical structures virtually generated are predicted by applying machine learning models to select chemical structures with predicted values satisfying the target. Because chemical structures within the applicability domain (AD) of prediction models tend to resemble those in the training dataset, the reliability of predicted values of y should be relatively high, though the y values for the structures within the AD tend to fall in a range of y values of the training dataset, making successful extrapolative design to be challenging. The aim of this study is to develop a path to successful extrapolative design by generating chemical structures near the AD boundary with their predicted values of y not falling in the range of y in the training dataset. Pareto-optimal selections were performed both for the chemical structures generated by a generative adversarial network (GAN) and those in its training data set, compromising the trade-off between the normalized distance from the AD boundary and the predicted value of y for a number of times equal to the number of training data points. This was done in an iterative manner. The proposed method was validated using six types of datasets and it was shown that the Pareto optimal solution of the GAN training data can successfully be updated by the method.

  • Data-Guided Exploration of BiVO4 Synthesis Conditions for Enhanced Photocatalytic Oxygen Evolution

    by Yuta Takami, Akihide Iwase, Hiromasa Kaneko

    BiVO4 is a visible-light-responsive oxygen-evolution photocatalyst whose activity depends strongly on synthesis-derived phase and microstructural characteristics. Here, Bayesian optimization (BO), used as an established sequential experimental-design method rather than as a newly proposed algorithm, was applied to explore liquid–solid phase BiVO4 synthesis conditions. Gaussian process regression related synthesis variables to oxygen evolution activity, and a genetic algorithm maximized acquisition functions to propose experiments. In a literature-initialized exploratory study, the highest observed operational activity index was 101.2 μmol h–1 g–1 L, and X-ray diffraction (XRD) indicated coexisting monoclinic and tetragonal BiVO4. In an experiment-initialized study conducted under a common evaluation protocol, all four proposed conditions exceeded the initial best observed activity of 40.5 μmol h–1, and the highest observed value was 52.0 μmol h–1; the corresponding sample was predominantly monoclinic by XRD. The best-performing conditions in the two studies shared nearly equimolar Bi and V precursor concentrations of approximately 0.19 M, HNO3 concentrations of 0.51–0.57 M, solution volumes near 51 mL, and prolonged room-temperature stirring. Because each condition was evaluated once, no statistical significance or experimental variance could be determined. Moreover, the literature-derived activity index does not correct for all differences in irradiation and reactor configuration, and no alternative-search benchmark or additional physicochemical characterization was performed. Accordingly, the results are interpreted as proof-of-concept evidence that sequential data-guided experimentation can identify promising BiVO4 synthesis conditions, while the mechanistic origin and comparative search efficiency remain subjects for future study.

  • Ranking-Based Surrogate Modeling for Bayesian Optimization under Small-Data Conditions

    by Yuya Endo, Hiromasa Kaneko

    Bayesian optimization (BO) is widely used for reaction optimization, but under small-data conditions, regression of absolute objective values may not align with the practical goal of prioritizing promising experiments. Here, we compared ranking-based BO (RankBO) combined with Thompson sampling (Rank-TS) with regression-based BO using two reaction benchmarks. Rank-TS showed a clear advantage on Direct Pd-catalyzed arylation in optimization performance, global ranking quality, and recovery of high-yielding conditions. For Suzuki–Miyaura coupling, which comprised 12 substrate combinations, its improvement was small in the pooled analysis. However, in the equal-weight macro analysis across the 12 substrate-defined search spaces, Rank-TS showed higher ranking quality and faster high-yield-condition recovery. These results indicate that Rank-TS is particularly useful for prioritizing reaction conditions within a fixed-substrate combination under small-data conditions, and that its relative performance depends on how the chemical search space and ranking task are defined.


  • Descriptor‐Based Zero‐Shot Learning for Predicting Catalytic Constituent Elements

    by Jun Maki, Hiromasa Kaneko

    Identifying effective combinations of constituent elements is a key challenge in catalyst design. However, experimentally evaluating all possible elemental combinations is impractical because of the combinatorial explosion. To address this issue, we investigated the application of descriptor‐based zero‐shot learning to catalyst design. Specifically, we examined whether the effects of elements absent from the training dataset could be predicted from the descriptors of other elements. A database of physicochemical properties of elements and their oxides was constructed and used to predict the performance of catalysts for the oxidative coupling of methane (OCM) reaction. The test data containing the target element were evaluated using 24 regression models, and the model yielding the highest coefficient of determination ( R 2 ) was selected. The results demonstrated that the catalytic contributions of elements such as La, Co, and Sr could be predicted with reasonable accuracy, suggesting the effectiveness of the proposed approach. In contrast, prediction for Li was considerably more difficult because the physicochemical properties of Li and its oxides differ substantially from those of other elements. These findings suggest that descriptor‐based zero‐shot learning is more effective for elements belonging to groups with similar physicochemical properties, such as transition metals and lanthanides.

  • Modular Composition-to-Structure Machine-Learning Workflow for Hypothesis Generation in Lithium Solid Electrolytes

    by Eri Ishikawa, Hiromasa Kaneko

    Machine-learning screening of lithium solid electrolytes is often performed either at the composition level, where polymorph-dependent transport cannot be represented, or within fixed crystal-structure databases, where unexplored compositions and structures are inaccessible. Here, we present a modular two-stage workflow that connects constrained composition generation and composition-based property prediction with stochastic crystal-structure generation, machine-learning-potential relaxation, and structure-aware ranking. The individual algorithms are established; the methodological contribution is their integration into a composition-to-structure hypothesis-generation pipeline that operates beyond a fixed list of known structures. The workflow was executed through M3GNet prerelaxation and internal ranking of 4600 PyXtal-generated trial structures. Because the structure-aware model has an MAE of 2.72 log10 units and no candidate-specific uncertainty analysis, applicability-domain analysis, DFT reoptimization, convex-hull analysis, transport simulation, or experimental validation was performed, the generated structure-level scores are not interpreted as quantitative conductivities or evidence of physical viability. The demonstrated contribution is therefore a validation-aware early triage architecture that produces hypotheses for subsequent high-fidelity assessment rather than a validated solid–electrolyte discovery.

  • Open Access

    Prediction of the optical properties and optimization of the manufacturing conditions of glass-ceramic products using machine learning

    by Makoto Takeuchi, Yusuke Hokii, Shigenori Akiyama, Takuya Sato, Hiromasa Kaneko

  • Design of Thermotropic Liquid Crystal Molecules With a Wide Temperature Range for the Liquid Crystal Phase Using Machine Learning

    by Naoki Masuyama, Hiromasa Kaneko

    Thermotropic liquid crystals are functional materials whose phase structures change with temperature. They exhibit mesophases—intermediate states between solid and liquid. There is demand for materials that maintain a stable mesophase over a wide temperature range Δ T LC , enabling operation even under harsh conditions. Conventional liquid crystal development requires significant time and money from molecular design to physical property evaluation. This study designs liquid crystal molecules exhibiting wide Δ T LC in a specific mesophase. The proposed method enables efficient liquid crystal exploration by predicting Δ T LC for new materials using machine learning. First, it identifies the mesophase formed from the molecular structure. Next, it predicts the phase transition temperatures (melting point and clearing point) for that phase. Inputting molecules with unknown Δ T LC into these models enables the prediction of Δ T LC for any given mesophase and efficient exploration of liquid crystal molecules.