Communication-Efficient Frank-Wolfe Algorithm for Nonconvex Decentralized Distributed Learning

Authors: Wenhan Xian, Feihu Huang, Heng Huang10405-10413

AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on matrix completion and model compression applications demonstrate the efficiency of our new algorithm.
Researcher Affiliation Collaboration Wenhan Xian 1, Feihu Huang 1, Heng Huang 1,2 1 Department of Electrical and Computer Engineering, University of Pittsburgh, Pittsburgh, USA 2 JD Finance America Corporation, Mountain View, CA, USA
Pseudocode Yes Algorithm 1 Decentralized Quantized Stochastic Frank Wolfe (DQSFW)
Open Source Code No The paper does not contain any statement about making its source code available or provide a link to a code repository.
Open Datasets Yes We run our experiment on two benchmark datasets, Movie Lens 100k and Movie Lens 1m (Harper and Konstan 2015).
Dataset Splits No For Movie Lens datasets, the paper states, 'we take all data for training'. For CIFAR-10, it mentions '50000 training samples' but does not specify a validation split or percentages for train/test/validation.
Hardware Specification Yes Each node is an Intel Xeon E5-2660 machine within an infiniband network. ... The experiment is implemented on 8 GTX1080 GPUs by Pytorch.
Software Dependencies No The paper mentions 'mpi4py' and 'Pytorch' as software used, but it does not specify version numbers for these dependencies.
Experiment Setup Yes For all of the three algorithms, we set step size ηt = t 0.75. ... For all three algorithms, step size is chosen as ηt = 1 2t 0.75. ... we use cross-entropy loss function as the criterion and set τ = 0.8.