Tuesday, November 17, 2026Latest
Michigan News

Michigan Tech researchers find AI needs less data to predict new materials

Researchers at Michigan Technological University have discovered that artificial intelligence models require significantly less data to accurately predict the electronic structures of novel materials. Led by Associate Professor Susanta Ghosh, the team found that AI models could discard 90% of their training data with virtually no loss in accuracy.

The study, published in npj Computational Materials, showed that even after removing 99% of the training data, predictions remained highly accurate while training times dropped by at least threefold. The findings apply to a wide range of materials, including metals, water, and complex alloys.

Ghosh attributes the efficiency to a minimal dataset carefully selected to capture data diversity, challenging the need for extensive datasets. The research, conducted in collaboration with UCLA and supported by the U.S. Department of Energy and National Science Foundation, could accelerate AI-driven discovery in materials science.

Read the full story at mtu.edu.

Leave a Reply