On the afternoon of December 15, Researcher Xu Changsheng from the Institute of Automation, Chinese Academy of Sciences, delivered an academic lecture to faculty and students at our college. The session was chaired by Dean Mao Qirong and attended by teachers and graduate students from related research fields.

The lecture focused on the theme “Research and Application of Multimodal Large Models in an Open World.” Researcher Xu addressed two key scientific questions: how to model multimodal correlations while preserving the distinct characteristics of each data type, and how to achieve cross-domain and few-shot transfer in downstream tasks without compromising model generalization. He analyzed the essential components of an ideal visual system and introduced Libra—a multimodal large model based on a decoupled visual architecture developed by his team. The talk also covered a series of methods designed to maintain generalization capability during cross-domain and few-shot transfer in multimodal large models.

Following the presentation, Researcher Xu engaged in an in-depth discussion with the audience, thoughtfully answering questions on multimodal learning, model transfer strategies, and practical application scenarios. His well-structured and insightful presentation, characterized by both rigor and innovation, provided valuable inspiration to faculty and students exploring cutting-edge developments in artificial intelligence.