Fast Algorithm for Constrained Linear Inverse Problems
Authors: Mohammed Rayyan Sheriff, Floor Fenne Redel, Peyman Mohajerin Esfahani
JMLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate the performance of FLIPS on the classical problems of Binary Selection, Compressed Sensing, and Image Denoising. We also provide open source MATLAB and PYTHON package for these three examples, which can be easily adapted to other LIPs. ... Numerical Results We test FLIPS on a few well-known LIPs namely Compressed Sensing, Binary-Selection problem, and finally we test FLIPS on the classical Image Denoising problem. |
| Researcher Affiliation | Academia | Mohammed Rayyan Sheriff EMAIL Floor Fenne Redel EMAIL Peyman Mohajerin Esfahani EMAIL Delft Center for Systems & Control Delft University of Technology Delft, The Netherlands |
| Pseudocode | Yes | Algorithm 1: The Fast LIP Solver |
| Open Source Code | Yes | We also provide open source MATLAB and PYTHON package for these three examples, which can be easily adapted to other LIPs. ... Associated with this algorithm, we also present an open-source Matlab package that includes the proposed algorithm (and also the implementation of CP and C-SALSA) (Sheriffet al., 2022). The open-source code can be found on the author s Github page (Sheriffet al., 2022). |
| Open Datasets | Yes | We demonstrate the performance of FLIPS on the classical problems of Binary Selection, Compressed Sensing, and Image Denoising. ... We consider the Compressed Sensing problem (34) for the standard images of Lena , Cameraman , and Barbara , each of size K = 64x64 pixels. ... Finally, we also consider another image processing task of denoising an image... We consider the three images: Lena , Barbara , and Cameraman , each of size 256x256. |
| Dataset Splits | No | The paper does not explicitly provide training/test/validation dataset splits. It describes data generation methods for some experiments (e.g., random vectors, noise) and patch-based processing for image tasks, but no standard dataset splits. |
| Hardware Specification | Yes | All experiments in Sections 4.2, 4.1, and 4.3 were run on a laptop with Apple M1 processor with 8GB RAM using MATLAB 2021b. While, the experiments in Section 4.4 comparing FLIPS with FISTA were run on a laptop with Apple M1 max processor with 32GB RAM using Python. |
| Software Dependencies | No | All experiments in Sections 4.2, 4.1, and 4.3 were run on a laptop with Apple M1 processor with 8GB RAM using MATLAB 2021b. While, the experiments in Section 4.4 comparing FLIPS with FISTA were run on a laptop with Apple M1 max processor with 32GB RAM using Python. |
| Experiment Setup | Yes | We chose the value of ϵ = 10σ√m. ... We apply FLIPS for each of the problems using an AQO oracle with parameters β and ρ tuned for faster convergence. ... For an image I, ... we collect the linear measurement xi, of the image I as xi = cJi I + wi; where ci ∈ RK is a random vector whose each entry is drawn uniformly from [-0.5, 0.5], and wi is the measurement noise drawn randomly from a Gaussian distribution with variance σ^2 = 0.0055. ... The parameter values in the Chambolle-Pock algorithm, C-SALSA, and the AQO oracle parameters in FLIPS are tuned to get the best results for each patch size. |