The field of artificial intelligence is witnessing significant advancements in multimodal and continual learning. Researchers are exploring new approaches to integrate multiple modalities, such as vision, language, and human motion, to enhance model performance and generalizability. One notable direction is the development of routing strategies that enable models to dynamically allocate experts or parameters based on input prompts or tasks, reducing catastrophic forgetting and improving adaptation to new domains. Another area of focus is the design of novel position encoding frameworks that preserve the inherent structures of each modality, leading to improved performance in vision-language models. Furthermore, researchers are investigating knowledge-guided prompt learning frameworks that leverage structured knowledge bases to enhance semantic representations and support reasoning in cross-domain recommendation tasks. Noteworthy papers in this area include Soft Task-Aware Routing of Experts for Equivariant Representation Learning, which introduces a routing strategy for projection heads that models them as experts, and KGBridge, a knowledge-guided prompt learning framework for cross-domain sequential recommendation. Additionally, papers like GNN-MoE and RoME demonstrate the effectiveness of graph-based contextual routing and domain-robust mixture-of-experts frameworks in achieving state-of-the-art performance in domain generalization and MILP solution prediction tasks.