Learning a Compressed Sensing Measurement Matrix via Gradient Unrolling

Authors: Shanshan Wu, Alex Dimakis, Sujay Sanghavi, Felix Yu, Daniel Holtmann-Rice, Dmitry Storcheus, Afshin Rostamizadeh, Sanjiv Kumar

ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We compare the empirical performance of 10 algorithms over 6 sparse datasets (3 synthetic and 3 real). Our experiments show that there is indeed additional structure beyond sparsity in the real datasets; our method is able to discover it and exploit it to create excellent reconstructions with fewer measurements (by a factor of 1.1-3x) compared to the previous state-of-the-art methods.
Researcher Affiliation Collaboration 1Department of Electrical and Computer Engineering, University of Texas at Austin, USA 2Google Research, New York, USA.
Pseudocode No The paper describes algorithmic steps and equations, but does not present them in a clearly labeled "Pseudocode" or "Algorithm" block.
Open Source Code Yes We implement ℓ1-AE in Tensorflow. Our code is available online: https://github.com/wushanshan/L1AE.
Open Datasets Yes Our second dataset Wiki10-31K is a multi-label dataset downloaded from this repository (Bhatia et al., 2017). We only use the label vectors to train our autoencoder. Our third dataset is RCV1(Lewis et al., 2004), a popular text dataset.
Dataset Splits Yes Table 1. Summary of the datasets. The validation set is used for parameter tuning and early stopping. ... Train / Valid / Test Size
Hardware Specification Yes A single NVIDIA Quadro P5000 GPU is used in the experiments.
Software Dependencies No The paper mentions "Tensorflow" and "Gurobi (a commercial optimization solver)" but does not provide specific version numbers for these software dependencies.
Experiment Setup Yes We use stochastic gradient descent to train the autoencoder. Before training, every sample is normalized to have unit ℓ2 norm. The parameters are initialized as follows: A Rm d is a random Gaussian matrix with standard deviation 1/ d; β is initialized as 1.0. Other hyperparameters are given in Appendix B. A single NVIDIA Quadro P5000 GPU is used in the experiments. We set the decoder depth T = 10 for most datasets.