Latest News
WeChat Official Account When the amount of data is small, the performance of deep learning will be greatly limited. Few-shot learning aims to use prior knowledge to quickly draw conclusions in limited new tasks, thus significantly narrowing the gap between artificial intelligence and humans. Centering around the theory of few-shot learning and focusing on knowledge representation methods as well as the embedding of knowledge in neural networks, we have studied the algorithm designs such as superclass representation, graph network models, knowledge reasoning, and zero-shot classification guided by CLIP. These efforts are aimed at alleviating common issues like catastrophic forgetting and feature drift during the process of incremental few-shot learning, and solving classic visual problems such as image semantic segmentation and image classification in the context of few-shot scenarios.
The special issue “Advances in Few-Shot Learning with Multimodal Large Models,” co-organized by Professor Wu Yirui from our lab and Professor Wan Shaohua from the University of Electronic Science and Technology of China, has been launched on Applied Sciences. The submission deadline is March 20, 2025. For more details, please visit https://www.mdpi.com/journal/applsci/special_issues/3N827RPJSC. We warmly invite all faculty and students to submit their papers.