One-sided Frank-Wolfe algorithms for saddle problems
Authors: Vladimir Kolmogorov, Thomas Pock
ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We show applications to labeling problems frequently appearing in machine learning and computer vision. and Preliminary numerical results are given in Section 5. and In this section, we show preliminary results for solving MRFs arising from computer vision. |
| Researcher Affiliation | Academia | 1Institute of Science and Technology Austria 2Institute of Computer Graphics and Vision, Graz University of Technology. Correspondence to: Vladimir Kolmogorov <vnk@ist.ac.at>, Thomas Pock <pock@icg.tugraz.at>. |
| Pseudocode | Yes | Algorithm 1 Algorithm FWε(x; g P). Output: vector x ε arg minx g P(x). and Algorithm 2 Approx. accelerated proximal gradient method and Algorithm 3 Inexact primal-dual algorithm. |
| Open Source Code | No | No statement providing concrete access to source code for the methodology described in this paper was found. |
| Open Datasets | Yes | Image denoising The first example is a simple image denoising problem... Stereo The second example is a classical disparity estimation problem from a rectified stereo image pair (Scharstein et al., 2014). |
| Dataset Splits | No | The paper describes input images for the problems (e.g., '150 200 pixels', '718 496 pixels') and the number of labels, but does not specify how the data was split into training, validation, and test sets. It mentions 'noisy input image' and 'left input image' as the basis for experiments, which implies evaluation on specific instances rather than a dataset split for reproducibility. |
| Hardware Specification | No | No specific hardware details (GPU/CPU models, memory, etc.) used for running experiments were mentioned in the paper. |
| Software Dependencies | No | The paper mentions algorithmic components and external tools (e.g., 'CNN-based correlation network'), but does not provide specific software dependencies with version numbers (e.g., Python 3.x, PyTorch 1.x, specific library versions). |
| Experiment Setup | Yes | We compare two versions of Algorithm 2: accelerated proximal point algorithm (A-PPA) with the aggressive choice tn = (n + 1)/2 ... and the standard proximal point algorithm (PPA) with the tn = 1. ... We invoke Algorithm 1 to minimize the functions Fγ, y up to accuracy εn = gap0 n α for a constant α > 0... |