Machine learning models for materials discovery face critical design flaw

A new research paper published on arXiv highlights a critical design choice in machine learning models used for materials discovery. The study demonstrates that whether a model predicts physically impossible properties, such as a non-zero piezoelectric tensor for centrosymmetric crystals, is determined by a single design bit related to parity labels. Models with parity labels achieved near-perfect accuracy by predicting zero for forbidden properties, while models without this label made physically impossible predictions for over 90% of tested crystals. AI

IMPACT Highlights a critical flaw in ML models for materials science, potentially impacting accuracy and reliability in scientific discovery.

RANK_REASON Research paper detailing a specific finding about machine learning model design. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• alphaXiv
• arXiv
• CatalyzeX
• Centrosymmetric crystals of biomolecules: the racemate method.
• DagsHub
• Gotit.pub
• Hugging Face
• IArxiv
• Influence Flower
• machine learning
• materials science
• Piezoelectric Tensor of Collagen Fibrils Determined at the Nanoscale
• ScienceCast

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →