Local Smoothness in Variance Reduced Optimization

Authors: Daniel Vainsencher, Han Liu, Tong Zhang

NeurIPS 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Empirically, we provide thorough numerical results to back up our theory. Additionally we present algorithms exploiting local smoothness in more aggressive ways, which perform even better in practice.
Researcher Affiliation Academia Daniel Vainsencher, Han Liu Tong Zhang Dept. of Operations Research & Financial Engineering Dept. of Statistics Princeton University Rutgers University Princeton, NJ 08544 Piscataway, NJ, 08854 {daniel.vainsencher,han.liu}@princeton.edu tzhang@stat.rutgers.edu
Pseudocode Yes Algorithm 1 Local SVRG is an application of Prox SVRG with w dependent regularization. ... Algorithm 2 Affine-SDCA: adapting to locally affine φi, with speedup approximately A (r). ... Pseudo code for the slightly long Algorithm 4 is in the supplementary material for space reasons.
Open Source Code No The paper does not provide any concrete access to source code (no specific repository link, explicit code release statement, or code in supplementary materials) for the methodology described.
Open Datasets Yes SVRG solving smoothed hinge loss SVM on MNIST 0/1. ... SDCA solving smoothed hinge loss SVM on Mushroom. ... SDCA solving smoothed hinge loss SVM on w8a. ... SDCA solving smoothed hinge loss SVM on Dorothea. ... SDCA solving smoothed hinge loss SVM on ijcnn1.
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 does not provide specific hardware details (exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running its experiments.
Software Dependencies No The paper does not provide specific ancillary software details (e.g., library or solver names with version numbers) needed to replicate the experiment.
Experiment Setup Yes On the left we see variants of SVRG with η = 1/ (8L)... The sampling distribution, step size and number of iterations in the latter are determined by smoothness of the losses.