Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1]

On the Influence of Momentum Acceleration on Online Learning

Authors: Kun Yuan, Bicheng Ying, Ali H. Sayed

JMLR 2016 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental From simulations, the equivalence between momentum and standard stochastic gradient methods is also observed for non-differentiable and non-convex problems. ... In this section we illustrate the main conclusions by means of computer simulations for both cases of mean-square-error designs and logistic regression designs.
Researcher Affiliation Academia Kun Yuan EMAIL Bicheng Ying EMAIL Ali H. Sayed EMAIL Department of Electrical Engineering University of California Los Angeles, CA 90095, USA
Pseudocode No The paper describes algorithms using mathematical recursions like (2) and (22)-(23) but does not present them in structured pseudocode blocks or figures, nor are there any sections explicitly labeled 'Pseudocode' or 'Algorithm'.
Open Source Code No The paper does not contain any explicit statement about releasing source code for the methodology described, nor does it provide a link to a code repository.
Open Datasets Yes In Section 7.4 'Visual Recognition', the paper states: 'We employ the CIFAR-10 database'. In Section 7.2 'Regularized Logistic Regression', it mentions 'a benchmark data set the Adult Data Set' and provides URLs: 'https://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/ or http://archive.ics.uci.edu/ml/datasets/Adult'.
Dataset Splits Yes In Section 7.2, for the Adult Data Set, it states: 'The set is divided into 6414 training data and 26147 test data'. In Section 7.4, for CIFAR-10, it states: 'There are 50000 training images and 10000 test images'.
Hardware Specification No The paper mentions 'computer simulations' and 'mini-batch stochastic-gradient learning' but does not provide specific details about the hardware (e.g., GPU models, CPU types, or memory) used for running the experiments.
Software Dependencies No The paper does not provide specific ancillary software details with version numbers (e.g., library or solver names with version numbers) needed to replicate the experiment.
Experiment Setup Yes Section 7.1 details for LMS: 'µ = µm = 0.003. The momentum parameter β is set as 0.9. ... µm = µ(1 β) = 0.0003.' Section 7.2 for Logistic Regression: 'µ = µm = 0.005. The momentum parameter β is set to 0.9. ... µm = µ(1 β) = 0.0005.' Section 7.4 for Neural Networks provides: 'ℓ2 regularization term is set to 0.001, initial value w-1 is generated by a Gaussian distribution with 0.05 standard deviation... batch size equal to 100... momentum parameter is set to β = 0.9, and the initial step-size µm is set to 0.01... reduce µm to 0.95µm after every epoch.' Similar details are given for the convolutional neural network, including batch size, step-size, and momentum parameters.