Dimension-Free Bounds for Low-Precision Training

Authors: Zheng Li, Christopher M. De Sa

NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments Next, we validate our theoretical results experimentally on convex problems. To do this, we analyzed how the size of the noise floor of convergence of SGD and LP-SGD varies as the dimension is changed for a class of synthetic problems.
Researcher Affiliation Academia Zheng Li IIIS, Tsinghua University lzlz19971997@gmail.com Christopher De Sa Cornell University cdesa@cs.cornell.edu
Pseudocode Yes Algorithm 1 LP-SGD: Low-Precision Stochastic Gradient Descent
Open Source Code No The paper does not contain any statement about making the source code for the described methodology publicly available, nor does it provide any links to a code repository.
Open Datasets Yes Second, we ran LP-SGD on the MNIST dataset [10].
Dataset Splits No The paper mentions using a synthetic dataset and the MNIST dataset, and describes some experimental parameters and running for 500 epochs, but does not explicitly provide information on train/validation/test splits (e.g., percentages, counts, or specific split files) for either dataset.
Hardware Specification No The paper does not provide any specific details about the hardware (e.g., CPU, GPU models, memory) used to conduct the experiments.
Software Dependencies No The paper does not provide specific version numbers for any software dependencies or libraries used in the experiments.
Experiment Setup Yes We set α = 0.01, β = 0.2, p1 = 0.9, pd = 0.001, and s = 16, we chose each entry of w uniformly from [-1/2, 1/2], and we set δ such that the low-precision numbers would range from -1 to 1.