PgM segments multimodal representations into uni-modal and paired-modal features with cumulative-softmax gates and trains them with separate learners, reconstruction, and classification losses, yielding accuracy gains over the simple baselines tested.
Improving multimodal fusion with hierarchical mutual information maximization for multimodal sentiment analysis
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Partitioner Guided Modal Learning Framework
PgM segments multimodal representations into uni-modal and paired-modal features with cumulative-softmax gates and trains them with separate learners, reconstruction, and classification losses, yielding accuracy gains over the simple baselines tested.