Estimation in Rotationally Invariant Generalized Linear Models via Approximate Message Passing

Authors: Ramji Venkataramanan, Kevin Kögler, Marco Mondelli

ICML 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Numerical results showcase a performance close to Vector AMP (which is conjectured to be Bayes-optimal in some settings), but obtained with a much lower complexity, as the proposed algorithm does not require a computationally expensive singular value decomposition.
Researcher Affiliation Academia 1University of Cambridge, United Kingdom 2ISTA, Austria.
Pseudocode No The algorithm steps are described textually and through equations (4)-(10), but there is no distinct block explicitly labeled 'Algorithm' or 'Pseudocode'.
Open Source Code No No explicit statement about providing source code for the proposed RI-GAMP method was found. The paper only refers to code for VAMP, a baseline method, available at 'https://sourceforge.net/projects/gampmatlab/'.
Open Datasets No The paper refers to 'synthetic data' and 'images' which include 'the sparse grayscale image considered in (Schniter & Rangan, 2014)' and 'the Haar wavelet transform of the RGB image in Figure 8a'. However, no concrete access information (e.g., specific link, DOI, or formal citation with authors and year for a publicly available dataset) is provided for these images or synthetic data generation parameters.
Dataset Splits No The paper mentions running experiments for '10 independent runs' but does not specify any training, validation, or test dataset splits or partitioning methodologies.
Hardware Specification No No specific hardware details (e.g., GPU/CPU models, memory, or cloud instance types) used for running the experiments are mentioned in the paper.
Software Dependencies No The paper mentions 'Pywavelets: A python package for wavelet analysis' and references MATLAB in the context of VAMP implementation, but does not provide specific version numbers for any software dependencies required for reproduction.
Experiment Setup Yes We implement the RI-GAMP given in (4)-(5), with initialization s1 = y and x1 = ATs1. The denoisers ft and ht+1, for t 1, are given by (28)-(29). The parameter λ is taken to be 1/6, which is close to the actual sparsity of the image given by 8645/50625; the parameter σ2 is taken to be 1/λ, which gives E{X2 } = 1.