An Accelerated Decentralized Stochastic Proximal Algorithm for Finite Sums

Authors: Hadrien Hendrikx, Francis Bach, Laurent Massoulié

NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we illustrate the theoretical results by showing how ADFS compares with MSDA [Scaman et al., 2017], ESDACD [Hendrikx et al., 2019], Point-SAGA [Defazio, 2016], and DSBA [Shen et al., 2018]. All algorithms (except for DSBA, for which we fine-tuned the step-size) were run with out-of-the-box hyperparameters given by theory on data extracted from the standard Higgs, Covtype and RCV1 datasets from Lib SVM.
Researcher Affiliation Academia Hadrien Hendrikx INRIA DIENS PSL Research University hadrien.hendrikx@inria.fr Francis Bach INRIA DIENS PSL Research University francis.bach@inria.fr Laurent Massouli e INRIA DIENS PSL Research University laurent.massoulie@inria.fr
Pseudocode Yes Algorithm 1 ADFS(A, (σi), (Li,j), (µk ), (pk ), ρ)
Open Source Code Yes A Python implementation of ADFS is also provided in supplementary material.
Open Datasets Yes data extracted from the standard Higgs, Covtype and RCV1 datasets from Lib SVM.
Dataset Splits No The paper mentions using standard datasets (Higgs, Covtype, RCV1) but does not provide specific details on how these datasets were split into training, validation, or test sets for their experiments.
Hardware Specification No Experiments were run in a distributed manner on an actual computing cluster. This does not provide specific hardware models or detailed specifications.
Software Dependencies No A Python implementation of ADFS is also provided in supplementary material. This mentions Python but no specific version or other software dependencies with version numbers.
Experiment Setup Yes All algorithms (except for DSBA, for which we fine-tuned the step-size) were run with out-of-the-box hyperparameters given by theory and logistic regression task with m = 104 points per node, regularization parameter σ = 1 and communication delays τ = 5 on 2D grid networks of different sizes.