Recovery Algorithm Design for Generalized Linear Models via Standard Approximate Bayesian Inference


In this talk, recovery algorithm design for generalized linear models (GLMs) using standard approximate Bayesian inference algorithms (approximate message passing (AMP), vector approximate message passing (VAMP), sparse Bayesian learning (SBL)) will be presented. Substantial examples such as image classification, parameter estimation from quantized data and phase retrieval can be formulated as a GLM problem. Compared to the standard linear models (SLMs), solving the GLMs is more challenging because of the coupling of the linear and nonlinear transforms. It is known that generalized approximate message passing (GAMP) algorithm has been proposed to solve the GLMs, while the relationship between AMP and GAMP may be unclear. Here we provide some insights into the relationship between the SLMs and GLMs. According to expectation propagation (EP), the GLM can be iteratively approximated as a sequence of SLM subproblems, and thus the standard Bayesian algorithm can be easily extended to solve the GLMs. This talk is based on joint work with Xiangming Meng and Sheng Wu.

Speaker: Dr Jiang Zhu, Ocean College, Zhejiang University (China)  

Short biography: Dr. Zhu received his B.E. and Phd degrees from Harbin Engineering University and Tsinghua University in 2011 and 2016, respectively. He has been a visiting student in Lehigh University from Feb. to Aug. 2015. Since June 2016, he has been a lecturer in Ocean College, Zhejiang University. He is a member of the Chinese Institute of Electronics (CIE) and IEEE. His research interests include Bayesian algorithm design for the GLM, line spectrum estimation problem, signal detection and estimation, unlabelled sensing, etc.

  Date and Time




  • University of Wollongong
  • University of Wollongong, New South Wales
  • Australia 2522
  • Building: 20
  • Room Number: 1