Variance Reduction via Primal-Dual Accelerated Dual Averaging for Nonsmooth Convex Finite-Sums

Authors: Chaobing Song, Stephen J Wright, Jelena Diakonikolas

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

Reproducibility Variable Result LLM Response
Research Type Experimental Numerical experiments reveal competitive performance of VRPDA2 compared to state-of-the-art approaches.
Researcher Affiliation Academia 1Department of Computer Sciences, University of Wisconsin Madison, Madison, WI.
Pseudocode Yes Algorithm 1 Primal-Dual Accelerated Dual Averaging (PDA2)
Open Source Code No The paper does not provide an explicit statement or link for the open-sourcing of its own methodology's code.
Open Datasets Yes We compare VRPDA2 with two competitive algorithms SPDHG (Chambolle et al., 2018) and PURE CD (Alacaoglu et al., 2020) on standard a9a and MNIST datasets from the LIBSVM library (LIB).1 Both datasets are large, with n = 32, 561, d = 123 for a9a, and n = 60, 000, d = 780 for MNIST. [Footnote 1: LIBSVM Library. https://www.csie.ntu.edu. tw/ cjlin/libsvm/index.html. Accessed: Feb. 3, 2020.]
Dataset Splits No The paper does not explicitly provide training/validation/test dataset splits needed to reproduce the experiment.
Hardware Specification No The paper does not explicitly describe the hardware used to run its experiments, such as specific GPU or CPU models.
Software Dependencies No The paper mentions the 'LIBSVM library' but does not specify its version number or other key software components with their versions.
Experiment Setup Yes For simplicity, we normalize each data sample to unit Euclidean norm, so that the Lipschitz constants appearing in the analysis (such as R0 in VRPDA2) are at most 1. We then scale these Lipschitz constants by {0.1, 0.25, 0.5, 0.75, 1}2... We fix the 1-regularization parameter λ to 10−4 and vary σ ∈ {0, 10−8, 10−4}, to represent the general convex, ill-conditioned strongly convex, and well-conditioned strongly convex settings, respectively.