<ul><li>Multi-Modal Learning (MML) aims to integrate information from diverse modalities for better predictive accuracy.</li><li>Existing methods aggregate gradients with fixed weights, overlooking the gradient uncertainty of each modality.</li><li>BOGC-MML is introduced as a Bayesian-Oriented Gradient Calibration method to address gradient uncertainties and optimize model direction.</li><li>The method models each modality's gradient as a random variable, quantifies uncertainties, and balances sensitivity and conservatism across dimensions.</li></ul>

Multi-Modal Learning with Bayesian-Oriented Gradient Calibration

Discover more