Lipschitz-Adaptive Alternating Minimization for robust multimodal sensor fusion

Key Points

• The aim is to enhance multimodal sensor fusion through a novel optimization framework.
• Proved marginal convexity for multimodal autoencoders.
• Developed a block-wise optimization strategy for multimodal architectures.
• Created an adaptive learning rate strategy based on Lipschitz conditions.
• Demonstrated improved performance on real-world heterogeneous multimodal datasets.
• Achieved robustness against scale mismatch in sensor data.