Short and Sparse Deconvolution --- A Geometric Approach

Authors: Yenson Lau, Qing Qu, Han-Wen Kuo, Pengcheng Zhou, Yuqian Zhang, John Wright

ICLR 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental 4 EXPERIMENTS
Researcher Affiliation Academia Yenson Lau Electrical Engineering Columbia University y.lau@columbia.edu Qing Qu Center for Data Science New York University qq213@nyu.edu Han-wen Kuo Electrical Engineering Columbia University hk2673@columbia.edu Pengcheng Zhou Department of Statistics Columbia University zhoupc2018@gmail.com Yuqian Zhang Electrical & Computer Engineering Rutgers University yqz.zhang@rutgers.edu John Wright Electrical Engineering Columbia University jw2966@columbia.edu
Pseudocode Yes Algorithm 1 Inertial Alternating Descent Method (i ADM) and Algorithm 2 Sa S-BD with homotopy continuation
Open Source Code Yes The code for implementations of our algorithms can be found online: https://github.com/qingqu06/sparse_deconvolution.
Open Datasets Yes Next, we test our method on real data10; Figures 5c and 5d demonstrate recovery of spike locations. 10Obtained at http://spikefinder.codeneuro.org. Here we will solve this task on the single-molecule localization microscopy (SMLM) benchmarking dataset11 via Sa SD... 11Data can be accessed at http://bigwww.epfl.ch/smlm/datasets/index.html. We consider frames video obtained via the two-photon calcium microscopy dataset from the Allen Institute for Brain Science13, shown in Figure 8. 13Obtained at http://observatory.brain-map.org/visualcoding/.
Dataset Splits No No explicit training/validation/test dataset splits (e.g., percentages, sample counts, or references to predefined splits) were found.
Hardware Specification No No specific hardware details (e.g., GPU/CPU models, memory, or cloud instance types) used for running the experiments were provided in the paper.
Software Dependencies No The paper does not provide specific software dependencies with version numbers (e.g., Python 3.x, PyTorch 1.x) that are needed to replicate the experiments.
Experiment Setup Yes For each method, we deconvolve signals with p0 50, m 100p0, and θ p 3{4 0 for both coherent and incoherent a0. For both i ADM, i ADM with homotopy, and i PALM we set α 0.3. For homotopy, we set λp1q maxℓ|xsℓrap0qs, yy|, λ 0.3 ?p0λ, and δ 0.5. Furthermore we set η 0.5 or η 0.8 and for ADMM, we set the slack parameter to ρ 0.7 or ρ 0.5 for incoherent and coherent a0 respectively. We set x0 i.i.d. Bernoullipp 4{5 0 q P R104 with additive noise n i.i.d. Np0, 5 10 2q. For experiments in the main text, we set ε max ! |x|prn{ logpm{nqsq , 10 3) .