Modality-Aware SAM: Sharpness-Aware-Minimization Driven Gradient Modulation for Harmonized Multimodal Learning

In multimodal learning, dominant modalities often overshadow others, limiting generalization. We propose Modality-Aware Sharpness-Aware Minimization (M-SAM), a model-agnostic framework that applies to many modalities and supports early and late fusion scenarios. In every iteration, M-SAM in three steps optimizes learning. $\mathbf{\text{First,}}\mathbf{\text{it}}\mathbf{\text{identifies}}\mathbf{\text{the}}\mathbf{\text{dominant}}\mathbf{\text{modality}}$ based on modalities’ contribution in the accuracy using Shapley. $\mathbf{\text{Second,}}\mathbf{\text{it}}\mathbf{\text{decomposes}}\mathbf{\text{the}}\mathbf{\text{loss}}\mathbf{\text{landscape}}$, or in another language, it modulates the loss to prioritize the robustness of the model in favor of the dominant modality, and $\mathbf{\text{third,}}\mathbf{\text{M-SAM}}\mathbf{\text{updates}}\mathbf{\text{the}}\mathbf{\text{weights}}$ by backpropagation of modulated gradients. This ensures robust learning for the dominant modality while enhancing contributions from others, allowing the model to explore and exploit complementary features that strengthen overall performance. Extensive experiments on four diverse datasets show that M-SAM outperforms the latest state-of-the-art optimization and gradient manipulation methods and significantly balances and improves multimodal learning.