A formula-free transformer with fuzzy MoE aggregation and contrastive alignment improves IR-to-SMILES elucidation, reaching 31.8% Top-1 on experimental NIST spectra, but the >10-point gains occur only on the QM9S subset.
End-to-end multimodal structure elucidation from raw spectra combining contrastive learning and evolutionary algorithms.Nature Communications, 17(5013), 2026
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Data Fusion and Contrastive Alignment for Unconstrained IR Molecular Structure Elucidation
A formula-free transformer with fuzzy MoE aggregation and contrastive alignment improves IR-to-SMILES elucidation, reaching 31.8% Top-1 on experimental NIST spectra, but the >10-point gains occur only on the QM9S subset.