High-Dimensional Variance-Reduced Stochastic Gradient Expectation-Maximization Algorithm

Authors: Rongda Zhu, Lingxiao Wang, Chengxiang Zhai, Quanquan Gu

ICML 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental demonstrate the advantages of our algorithm by both theoretical analysis and numerical experiments. and In this section, we present experiment results to validate our theory.
Researcher Affiliation Collaboration 1Facebook, Inc., Menlo Park, CA 94025 2Department of Computer Science, University of Virginia, Charlottesville, VA 22904, USA 3Department of Computer Science, University of Illinois at Urbana-Champaign, Urbana, IL 61801.
Pseudocode Yes Algorithm 1 Variance Reduced Stochastic Gradient EM Algorithm (VRSGEM)
Open Source Code No The paper does not contain any statement or link indicating that the source code for the described methodology is publicly available.
Open Datasets No The paper describes generating synthetic data for experiments based on specified parameters (e.g., 'The covariance matrix Σ of V is chosen to be a diagonal matrix with all elements being 1. We randomly set two elements to λmax(Σ) = 10, and another two elements to λmin(Σ) = 0.1.'), but does not provide access information (link, DOI, or citation) to a publicly available dataset.
Dataset Splits No The paper describes overall experiment settings including sample sizes (N), but does not explicitly provide specific train/validation/test dataset splits (percentages or counts) or refer to standard predefined splits.
Hardware Specification No The paper does not provide specific details about the hardware (e.g., GPU/CPU models, memory) used to run the experiments.
Software Dependencies No The paper does not provide specific version numbers for any ancillary software dependencies (e.g., programming languages, libraries, or frameworks) used in the experiments.
Experiment Setup Yes All the comparisons are under two different parameter settings: s = 5, d = 256, b = 100, N = 5000 and s = 10, d = 512, b = 200, N = 10000. For VRSGEM, we choose m = 30, n = 50 and T = 50 across all settings and models. ... The learning rate η is tuned by grid search and s is chosen by cross validation. We use random initialization.