Developing machine learning for heterogeneous catalysis with experimental and computational data.

Area of Science:

• Catalysis
• Materials Science
• Computational Chemistry

Background:

• Machine learning (ML) excels at pattern recognition in large datasets, including catalyst performance prediction.
• Current ML models in catalysis primarily use high-throughput quantum chemistry, with limited experimental validation.
• Simplified computational models and scarce experimental data hinder ML's success in catalysis.

Purpose of the Study:

• To review studies integrating high-throughput methods and ML for solid heterogeneous catalysis.
• To analyze trends in ML model descriptors, materials, reactions, and dataset sizes.
• To assess ML model performance using R-squared values across different trends.

Main Methods:

• Systematic analysis of existing literature on ML in heterogeneous catalysis.
• Categorization of studies based on input/output descriptors, materials, reactions, and dataset size.
• Comparison of ML model performance (R-squared values) based on identified trends.

Main Results:

• Identified key trends in ML applications for heterogeneous catalysis.
• Highlighted the importance of integrating experimental and computational data.
• Provided a comparative analysis of model performances based on various factors.

Conclusions:

• Integrating high-throughput computations and ML holds significant promise for advancing heterogeneous catalysis.
• Addressing limitations in computational models and experimental data is crucial for broader ML adoption.
• This review offers insights into current trends and future directions for ML in catalyst design.