Home People Projects Publications Seminar Activities Data&Codes PAL组举办总结年会2021年2月5日

Research team overview

The Pattern Analysis and Learning (PAL) Team, associated with the State Key Laboratory of Multimodal Artificial Intelligence Systems (formerly the National Laboratory of Pattern Recognition, NLPR), was founded in 2005. The research topics of the team include the theory and methods of pattern recognition and machine learning, document analysis and understanding, multimodal artificial intelligence.

Pattern recognition theory and methods are at the core of artificial intelligence, with wide applications in the society. Last years have seen tremendous advances in deep learning (DL) methods and applications. Yet the DL-based methods still encounter various challenges, include the open world, curse of dimensionality, limited labeled data, large category set, changing distribution, multiple domains, multiple modalities, complex structure and context, etc. Large language models (LLMs) and multimodal large models (MLLMs) are advancing AI rapidly, and still facing new challenges. Besides more powerful neural network architectures, representation models and learning algorithms, fusing neural networks and symbolic system and incorporating various knowledge are receiving increasing attention in AI field.

Document analysis, recognition and understanding (currently called as Document Intelligence) has long been a highly attended area of AI and pattern recognition, since characters and documents are popular in everywhere and all time. Document analysis and recognition (DAR) has been advanced largely in recent years, and combing DAR and natural language processing (NLP) is emerging as a new trend for understanding the information in document images. The remaining challenges include the recognition of complicated documents, degraded and distorted documents, character structure interpretation, graphics and omni-elements recognition, explainability and confidence estimation, semantic information extraction, understanding and question answering. Representation learning, structured learning and multimodal large models (MLLMs) are playing increasing roles in document recognition and understanding.

Tasking advantage of expertise in the above topics, we are extending our research into more AI topics and applications, including mathematical reasoning and application in education, spatiotemporal prediction, AI application to science (AI for Science) and engineering.

》》》》》Research Group Introduction Slides (Download)