Fig. 5: The contribution for each modality across prediction tasks.
From: Unified multimodal multidomain polymer representation for property prediction
The heatmap visualizes the modality-specific attention weights learned by the Uni-Poly model for each prediction task.
