Geometric Computing and AI-driven Drug Discovery

Dr. Xu's current research focuses on the following directions: 1) High-dimensional Geometric Algorithms and Joint Influence Theory, 2) Deep Learning, especially Geometry Enhanced Deep Learning and Large Models, 3) AI-driven Drug Discovery, and 4) Differential Privacy. 

High-dimensional geometric algorithms study the properties of data in high-dimensional geometric spaces and leverage these properties to design optimal algorithms for various problems. Joint Influence theory is an emerging new theory. It focuses on a class of natural phenomena, where multiple entities jointly influence another entity, thereby altering its properties or behavior. This theory has found applications in multiple fields, including deep learning, fast pattern matching, and 3D geological modeling. Geometry-enhanced deep learning is a promising new research direction that utilizes geometric properties of deep neural networks to improve the performance of various learning models. Deep neural networks possess unique geometric properties. How to exploit such properties to enhance model performance has been a crucial research topic since the inception of artificial neural networks. AI-driven drug discovery refers to the use of artificial intelligence methods to aid in drug development and discovery. It represents a significant application area of current AI. Differential privacy is the most rigorous privacy protection method that has rapidly developed in recent years. It enables the secure use of large amounts of sensitive private data for model training. Developing differential privacy algorithms suitable for various models is an urgent and popular research problem in current artificial intelligence research.