Statistical analysis of stochastic gradient methods for generalized linear models

Authors: Panagiotis Toulis, Edoardo Airoldi, Jason Rennie

ICML 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our set of experiments confirm our theory and more broadly suggest that the implicit procedure can be a competitive choice for fitting large-scale models, especially when robustness is a concern. We illustrate the different aspects of our theory on three separate sets of experiments.
Researcher Affiliation Collaboration Department of Statistics, Harvard University, 1 Oxford Street, Cambridge, MA 02138, USA. Google, Inc., 1600 Amphitheatre Pkwy, Mountain View, CA 94043
Pseudocode Yes Algorithm 1 Implicit learning of canonical GLMs
Open Source Code Yes The full version of the paper together with the accompanying source code and documentation can be found at the following location: http://www.people.fas.harvard.edu/ptoulis/harvard-homepage/implicit-sgd.html.
Open Datasets Yes We implement an implicit online learning procedure for a SVM model and compare it to a standard SGD method on the RCV1 benchmark.
Dataset Splits No The paper mentions 'Test errors' on the RCV1 dataset but does not specify the train/validation/test splits, their percentages, or sample counts, or refer to standard splits with citations.
Hardware Specification No The paper does not specify any particular hardware components such as GPU or CPU models, memory, or specific computing platforms used for the experiments.
Software Dependencies No The paper mentions using 'Bottou’s SVM SGD implementation' and 'Our implicit SVM' but does not specify versions for any software, libraries, or frameworks used in the experiments.
Experiment Setup No The paper mentions learning rate schedules (e.g., an = α/n) and regularization parameters (λ), but it does not provide a comprehensive set of hyperparameters, optimizer settings, or other detailed training configurations necessary for reproducibility.