On the redundancy in large material datasets: efficient and robust learning with less data

Abstract

Extensive efforts to gather materials data have largely overlooked potential data redundancy. In this study, we present evidence of a significant degree of redundancy across multiple large datasets for various material properties, by revealing that up to 95 % of data can be safely removed from machine learning training with little impact on in-distribution prediction performance. The redundant data is related to over-represented material types and does not mitigate the severe performance degradation on out-of-distribution samples. In addition, we show that uncertainty-based active learning algorithms can construct much smaller but equally informative datasets. We discuss the effectiveness of informative data in improving prediction performance and robustness and provide insights into efficient data acquisition and machine learning training. This work challenges the "bigger is better" mentality and calls for attention to the information richness of materials data rather than a narrow emphasis on data volume.

arXiv commentMain text: 9 pages, 2 tables, 6 figures. Supplemental information: 31 pages, 1 table, 24 figures

References in corpus (19)

• XGBoost: A Scalable Tree Boosting System
• Recent Advances and Applications of Deep Learning Methods in Materials Science
• A Universal Graph Deep Learning Interatomic Potential for the Periodic Table
• Less is more: sampling chemical space with active learning
• The Open Catalyst 2020 (OC20) Dataset and Community Challenges
• Atomistic Line Graph Neural Network for Improved Materials Property Predictions
• On-the-fly Closed-loop Autonomous Materials Discovery via Bayesian Active Learning
• The Joint Automated Repository for Various Integrated Simulations (JARVIS) for data-driven materials design
• Opportunities and Challenges for Machine Learning in Materials Science
• Towards Universal Neural Network Potential for Material Discovery Applicable to Arbitrary Combination of 45 Elements
• On scientific understanding with artificial intelligence
• The Open Catalyst 2022 (OC22) Dataset and Challenges for Oxide Electrocatalysts
• High-Dimensional Materials and Process Optimization using Data-driven Experimental Design with Well-Calibrated Uncertainty Estimates
• A critical examination of robustness and generalizability of machine learning prediction of materials properties
• Unified Graph Neural Network Force-field for the Periodic Table
• On the redundancy in large material datasets: efficient and robust learning with less data
• GemNet-OC: Developing Graph Neural Networks for Large and Diverse Molecular Simulation Datasets
• ET-AL: Entropy-Targeted Active Learning for Bias Mitigation in Materials Data
• Dataset Pruning: Reducing Training Data by Examining Generalization Influence

Cited by in corpus (9)

• On the redundancy in large material datasets: efficient and robust learning with less data
• Probing out-of-distribution generalization in machine learning for materials
• JARVIS-Leaderboard: A Large Scale Benchmark of Materials Design Methods
• Model-free quantification of completeness, uncertainties, and outliers in atomistic machine learning using information theory
• Efficient first principles based modeling via machine learning: from simple representations to high entropy materials
• Emerging Microelectronic Materials by Design: Navigating Combinatorial Design Space with Scarce and Dispersed Data
• FibreCastML: An Open Web Platform for Predicting Electrospun Nanofibre Diameter Distributions
• Do Graph Neural Networks Work for High Entropy Alloys?
• Design of a specimen to train path-dependent deep learning material models from a single uniaxial test: eliciting strain diversity via automatically differentiable elastoplastic topology optimization