SLowcalSGD : Slow Query Points Improve Local-SGD for Stochastic Convex Optimization

Authors: Tehila Dahan, Kfir Y. Levy

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental To assess the effectiveness of our proposed approach, we conducted experiments on the MNIST [26] dataset a well-established benchmark in image classification comprising 70,000 grayscale images of handwritten digits (0 9), with 60,000 images designated for training and 10,000 for testing. The dataset was accessed via torchvision (version 0.16.2). We implemented a logistic regression model [7] using the Py Torch framework and executed all computations on an NVIDIA L40S GPU. To ensure robustness, results were averaged over three different random seeds.
Researcher Affiliation Academia Tehila Dahan Department of Electrical Engineering Technion Haifa, Israel t.dahan@campus.technion.ac.il Kfir Y. Levy Department of Electrical Engineering Technion Haifa, Israel kfirylevy@technion.ac.il
Pseudocode Yes Algorithm 1 Parallel Stochastic Optimization Template Algorithm 2 SLowcal-SGD
Open Source Code Yes The complete codebase for these experiments is publicly available on our Git Hub repository.4 4https://github.com/dahan198/slowcal-sgd
Open Datasets Yes To assess the effectiveness of our proposed approach, we conducted experiments on the MNIST [26] dataset a well-established benchmark in image classification comprising 70,000 grayscale images of handwritten digits (0 9), with 60,000 images designated for training and 10,000 for testing. The dataset was accessed via torchvision (version 0.16.2)....
Dataset Splits Yes The MNIST [26] dataset a well-established benchmark in image classification comprising 70,000 grayscale images of handwritten digits (0 9), with 60,000 images designated for training and 10,000 for testing. The dataset was accessed via torchvision (version 0.16.2).
Hardware Specification Yes We implemented a logistic regression model [7] using the Py Torch framework and executed all computations on an NVIDIA L40S GPU.
Software Dependencies Yes The dataset was accessed via torchvision (version 0.16.2). We implemented a logistic regression model [7] using the Py Torch framework and executed all computations on an NVIDIA L40S GPU.
Experiment Setup Yes For fairness, the learning rate was selected through grid search, with a value of 0.01 for SLowcal-SGD and Local-SGD, and 0.1 for Minibatch-SGD. More details about the data distribution across workers and complete experimental results are provided in Appendix M.