Composition plus simulated XRD, fused by cross-attention with a new masked-XRD pretraining objective, reaches near-structure-based property accuracy on 5M materials.
Align before fuse: Vision and language representation learning with momentum distillation
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
XxaCT-NN: Structure Agnostic Multimodal Learning for Materials Science
Composition plus simulated XRD, fused by cross-attention with a new masked-XRD pretraining objective, reaches near-structure-based property accuracy on 5M materials.