AECF claims a single lightweight fusion layer improves robustness and calibration under missing modalities, but the paper's own ablation and proofs undermine the claim.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
stat.ML 1years
2025 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
Robust Multimodal Learning via Entropy-Gated Contrastive Fusion
AECF claims a single lightweight fusion layer improves robustness and calibration under missing modalities, but the paper's own ablation and proofs undermine the claim.