Sihan Xu (Si.X)
Ph.D. Student
University of Michigan
Email: sihanxu [at] umich [dot] edu
Research
My work aspires to develop systems that learn meaning representations and generate consistent semantics autonomously with minimal supervision involved. My philosophy is to pursue minimal parameterizations that effectively represent the maximal information within the input data distribution. By embracing this vision, I design algorithms for
(1) self-supervised learning or learning with minimal semantic supervision and
(2) generate consistent semantics given multimodal conditioning, with a primary application to visual generation (for now).
Represenation Learning
Generative Algorithms
Others
Selected Publications
(* indicate equal contribution, † indicate correspondence or project lead.)
If you like my figures here, I highly recommend you also visit Ziqiao Ma's homepage.
