FR3.R9.4

Overflow-Avoiding Memory AMP

Shunqi Huang, Japan Advanced Institute of Science and Technology, Japan; Lei Liu, Zhejiang University, China; Brian Kurkoski, Japan Advanced Institute of Science and Technology, Japan

Session:
Signal Processing 1

Track:
12: Signal Processing

Location:
Lamda

Presentation Time:
Fri, 12 Jul, 15:35 - 15:55

Session Chair:
Anand Sarwate, Rutgers
Abstract
Approximate Message Passing (AMP) type algorithms are widely used to the signal recovery in high-dimensional noisy linear systems. Recently, a principle called Memory AMP (MAMP) was emerged. Leveraging this principle, a Bayes-optimal MAMP (GD-MAMP) algorithm was created, inheriting the strengths of AMP and OAMP/VAMP algorithms. In this paper, we first provide an overflow-avoiding GD-MAMP (OA-GD-MAMP) to address the overflow problem that arises from some intermediate variables exceeds the range of floating point numbers. Second, we develop a complexity-reduced GD-MAMP (CR=GD-MAMP) that effectively reduces the number of matrix-vector products per iteration by 1/3 (from 3 to 2) without significantly slowing down convergence speed.
Resources