Globally Convergent Parallel MAP LP Relaxation Solver using the Frank-Wolfe Algorithm

Authors: Alexander Schwing, Tamir Hazan, Marc Pollefeys, Raquel Urtasun

ICML 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our method proves superior when compared to existing algorithms on a set of spin-glass models and protein design tasks. and 4. Experimental Evaluation We compare our approach to a wide variety of state-of-the-art baselines using spin-glass models of size 10 10 with variable state-space size and energy functions arising from a protein design task.
Researcher Affiliation Academia Alexander G. Schwing ASCHWING@CS.TORONTO.EDU University of Toronto, 10 King s College Rd., Toronto, Canada Tamir Hazan TAMIR@CS.HAIFA.AC.IL University of Haifa, Haifa, Israel Marc Pollefeys MARC.POLLEFEYS@INF.ETHZ.CH ETH Zurich, Universit atstrasse 6, Zurich, Switzerland Raquel Urtasun URTASUN@CS.TORONTO.EDU University of Toronto, 10 King s College Rd., Toronto, Canada
Pseudocode Yes Figure 2. Frank-Wolfe algorithm for finding the solution of the program given in Eq. (4). and Figure 4. Our efficient, parallel and provably convergent MAP LP Relaxation Solver.
Open Source Code No The paper does not provide concrete access to source code for the methodology described, nor does it explicitly state that code will be released or is available.
Open Datasets Yes To this end, we make use of the eight problems from the probabilistic inference challenge1. (1http://www.cs.huji.ac.il/project/PASCAL/index.php)
Dataset Splits No The paper does not provide specific dataset split information (exact percentages, sample counts, citations to predefined splits, or detailed splitting methodology) needed to reproduce the data partitioning.
Hardware Specification No The paper mentions 'single core' and '16 cores' for parallelization but does not provide specific hardware details such as exact GPU/CPU models, processor types, or memory amounts.
Software Dependencies No The paper mentions using baselines from the 'STAIR library by Gould et al. (2011)' but does not provide specific version numbers for this or any other software dependencies.
Experiment Setup Yes All algorithms are restricted to at most 5,000 iterations and all baselines utilize a single core. ... We start from ϵ = 0.01 and successively decrease its value if the model is sufficiently close to |R|ϵ optimality, i.e., if ϵ is larger than f(b)/1000. ... In the following we set the number of it-erations to 400 ... we increase the number of the maximally possible iterations to 50,000 and the Frank-Wolfe iterations to 2,000