MO4.R9.3

A Convergence Analysis of Approximate Message Passing with Non-Separable Functions and Applications to Multi-Class Classification

Burak Çakmak, Technical University of Berlin, Germany; Yue M. Lu, Harvard University, United States; Manfred Opper, Technical University of Berlin, Germany
Session:
AMP, Sparsity and Sketching
Track:
11: Information Theory and Statistics
Location:
Lamda
Presentation Time:
Mon, 8 Jul, 17:05 - 17:25
Session Chair:
Ramji Venkataramanan, University of Cambridge
Abstract
Motivated by the recent application of approximate message passing (AMP) to the analysis of convex optimizations in multi-class classifications [Loureiro, et. al., 2021], we present a convergence analysis of AMP dynamics with non-separable multivariate nonlinearities. As an application, we present a complete (and independent) analysis of the motivated convex optimization problem.
Session MO4.R9
MO4.R9.1: On the Success Probability of the $L_0$-regularized Box-constrained Babai Point
Xiao-Wen Chang, Yingzi Xu, McGill University, Canada
MO4.R9.2: Linear Operator Approximate Message Passing: Power Method with Partial and Stochastic Updates
Riccardo Rossetti, Duke University, United States; Bobak Nazer, Boston University, United States; Galen Reeves, Duke University, United States
MO4.R9.3: A Convergence Analysis of Approximate Message Passing with Non-Separable Functions and Applications to Multi-Class Classification
Burak Çakmak, Technical University of Berlin, Germany; Yue M. Lu, Harvard University, United States; Manfred Opper, Technical University of Berlin, Germany
MO4.R9.4: Efficient Nonconvex Optimization for Two-way Sparse Reduced-Rank Regression
Cheng Cheng, Ziping Zhao, ShanghaiTech University, China
Resources
View Manuscript