A Continuous-Time Mirror Descent Approach to Sparse Phase Retrieval

Authors: Fan Wu, Patrick Rebeschini

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

Reproducibility Variable Result LLM Response
Research Type Experimental In the following, we present simulations showing how the parameter β affects the convergence of HWF. The setup for the simulations is as follows. We generate a 10-sparse signal vector x Rn with x min = Ω(1/√k) from e O(k2) Gaussian measurements, where e O hides logarithmic terms. To the best of our knowledge, this is the first result on continuous-time solvers for (sparse) phase retrieval. We sample m = 1000 Gaussian measurement vectors Aj N(0, I50000) and generate phaseless measurements Yj = (AT j x )2.
Researcher Affiliation Academia Fan Wu Department of Statistics University of Oxford fan.wu@stats.ox.ac.uk Patrick Rebeschini Department of Statistics University of Oxford patrick.rebeschini@stats.ox.ac.uk
Pseudocode No The paper provides mathematical equations describing the algorithms (e.g., equations 1, 11, 13, 14), but it does not present them as structured pseudocode or clearly labeled algorithm blocks.
Open Source Code No The paper does not include an explicit statement about releasing source code for the methodology described, nor does it provide a link to a code repository.
Open Datasets No The paper describes generating synthetic data for its simulations ('We generate a 10-sparse signal vector... We sample m = 1000 Gaussian measurement vectors...'), rather than using a publicly available dataset with concrete access information such as a URL, DOI, or specific repository name.
Dataset Splits No The paper describes data generation for simulations but does not specify train, validation, or test dataset splits, nor does it mention cross-validation or predefined splits with citations.
Hardware Specification No The paper does not provide any specific details about the hardware used for running the simulations, such as GPU/CPU models or memory specifications.
Software Dependencies No The paper does not mention any specific software dependencies with version numbers (e.g., programming languages, libraries, or solvers) that would be needed to replicate the experiments.
Experiment Setup Yes The setup for the simulations is as follows. We generate a 10-sparse signal vector x R50000 by first drawing x N(0, I50000), then setting 49990 random entries of x to zero, and finally normalizing the vector to x 2 = 1. We sample m = 1000 Gaussian measurement vectors Aj N(0, I50000) and generate phaseless measurements Yj = (AT j x )2. For the step size η, we follow [54] and choose η = 0.1.