<ul><li>Multimodal fusion is often affected by modality imbalance, which can lead to biased learning and suboptimal fusion.</li><li>A Shapley-guided alternating training framework has been proposed to prioritize minor modalities adaptively, balancing and enhancing fusion.</li><li>The method uses Shapley Value-based scheduling to improve training sequences, ensuring under-optimized modalities receive sufficient learning.</li><li>The proposed approach, evaluated across four benchmark datasets, achieves state-of-the-art results in balance and accuracy in multimodal learning.</li></ul>

Modality Equilibrium Matters: Minor-Modality-Aware Adaptive Alternating for Cross-Modal Memory Enhancement

Discover more