The field of multimodal learning is witnessing significant developments, with a focus on addressing challenges such as modality imbalance and missing modalities. Researchers are exploring innovative strategies, including unidirectional dynamic interaction and cross-modal prompt learning, to improve the performance of multimodal models. Information-theoretic approaches, such as balanced information bottlenecks and comprehensive multi-view learning frameworks, are also being proposed to advance the field. These approaches aim to learn representations that preserve essential label-related information and facilitate effective cross-modal feature learning. Noteworthy papers in this area include: Mixture of Balanced Information Bottlenecks for Long-Tailed Visual Recognition, which proposes a novel structure for learning sufficient representations, and Balanced Multimodal Learning: An Unidirectional Dynamic Interaction Perspective, which introduces a proactive, sequential training scheme to mitigate modality imbalance. Additionally, Towards Comprehensive Information-theoretic Multi-view Learning presents a framework that discards the assumption of multi-view redundancy, considering both common and unique information for predictive tasks. Other notable works, such as Resilient Multimodal Industrial Surface Defect Detection with Uncertain Sensors Availability and Robult: Leveraging Redundancy and Modality Specific Features for Robust Multimodal Learning, demonstrate the effectiveness of these approaches in various applications.