Near-term quantum machines provide a novel way to explore many scientfic domains beyond the reach of classical machines. Meanwhile, near-term quantum machines are fragile, where the available quantum resources are limited and error-prone. Variational quantum algorithms (VQAs) are leading candidates to alleviate these defects. Experimental studies have demonstrated the potential of VQAs in a plethora of areas including machine learning, fundamental science, and quantum chemisty. Neverthess, theoretical understanding of VQAs remains largely unknown. To address this issue, in this talk, we investigate the expressivity of VQAs through the lens of statistical learning theory. According to the entangled relation between expressivity and model power, we further utilize the achieved results to analyze the generalization ability of a wide class of quantum discriminative and generative learning models and discuss potential advantages.  

　　Bio 

　　Yuxuan Du is currently a Senior Researcher at JD Explore Academy and also a member of Doctor Management Trainee at JD. com, Inc. Prior to that, he received a Ph.D. degree in computer science from The University of Sydney and a Bachelor of Physics (elite class) from Sichuan University. His research interests include fundamental algorithms for quantum machine learning, quantum learning theory, and AI for quantum science. He has published his research outcomes in many top-tier journals and conferences in physics and computer science including Physical Review Letters, Physical Review X Quantum, npj Quantum Information, Transactions on Information Theory, Conference on Computer Vision and Pattern Recognition, etc.