Towards Sample-Optimal Compressive Phase Retrieval with Sparse and Generative Priors
Authors: Zhaoqiang Liu, Subhroshekhar Ghosh, Jonathan Scarlett
NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We perform synthetic experiments to complement our theoretical results. We focus only on the sparse setting, in part because numerous existing algorithms are known for sparse priors but not for generative priors. |
| Researcher Affiliation | Academia | Zhaoqiang Liu National University of Singapore dcslizha@nus.edu.sg Subhroshekhar Ghosh National University of Singapore subhrowork@gmail.com Jonathan Scarlett National University of Singapore scarlett@comp.nus.edu.sg |
| Pseudocode | Yes | Algorithm 1 A spectral initialization method for sparse phase retrieval based on SPCA (PRI-SPCA) |
| Open Source Code | No | The paper mentions a MATLAB package from [28] for comparison: "We use the MATLAB package shared by the authors of [28] at https://github.com/Gauri Jagatap/model-copram." However, this is for a third-party method and not for the code implementing the authors' own PRI-SPCA method. |
| Open Datasets | No | The paper conducts "synthetic experiments" and generates data internally: "The support of the signal vector x is uniformly random, and the non-zero entries are generated from i.i.d. standard normal distributions." It does not use or provide access to a publicly available or open dataset. |
| Dataset Splits | No | The paper conducts synthetic experiments where data is generated on demand, and results are averaged over random trials. It does not specify fixed training, validation, or test dataset splits in the conventional sense (e.g., percentages or sample counts for distinct partitions). |
| Hardware Specification | Yes | All numerical experiments are conducted using MATLAB R2014b on a machine with an Intel Core i5 CPU at 1.8 GHz and 8GB RAM. |
| Software Dependencies | Yes | All numerical experiments are conducted using MATLAB R2014b on a machine with an Intel Core i5 CPU at 1.8 GHz and 8GB RAM. |
| Experiment Setup | Yes | For PRI-SPCA, we set l = 1 and u = 5, as previously suggested in [72]. For GRQI, we set the deflation parameter to be 0.2, and the total number of iterations to be 100. The number of power method steps used in all the methods is fixed to be 100. For the initialization method used in Th WF, we set the tuning parameter as α = 0.1, as suggested in [7]. Similarly, we set |I| = 1/6m for the initialization method of SPARTA. The noise vector η is generated from the distribution N(0, σ2 x 2 2Im). |