OrthTD disentangles shared and task-specific subspaces via orthogonality in a multimodal Transformer, yielding 87.5% average AUC and 37.2% average AUPRC on 12,430 surgical patients, outperforming prior multi-task and tabular methods especially on imbalanced outcomes.
Lightgbm: A highly efficient gradient boosting decision tree
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Disentangling Shared and Task-Specific Representations from Multi-Modal Clinical Data
OrthTD disentangles shared and task-specific subspaces via orthogonality in a multimodal Transformer, yielding 87.5% average AUC and 37.2% average AUPRC on 12,430 surgical patients, outperforming prior multi-task and tabular methods especially on imbalanced outcomes.