Non-Gaussian Tensor Programs

Authors: Eugene Golikov, Greg Yang

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We empirically validate Corollaries 4.3 and 4.4 in Appendix N. If you ran experiments...Did you include the code, data, and instructions needed to reproduce the main experimental results (either in the supplemental material or as a URL)? [Yes] Did you report error bars (e.g., with respect to the random seed after running experiments multiple times)? [Yes] See plots in Appendix N.
Researcher Affiliation Collaboration Eugene Golikov Ecole Polytechnique F ed erale de Lausanne Lausanne, Switzerland evgenii.golikov@epfl.ch Greg Yang Microsoft Research Seattle, USA gregyang@microsoft.com
Pseudocode No The paper describes mathematical frameworks and theorems but does not include any pseudocode or clearly labeled algorithm blocks.
Open Source Code No The paper's reproducibility checklist states "Did you include the code, data, and instructions needed to reproduce the main experimental results (either in the supplemental material or as a URL)? [Yes]". However, the main body or appendices of the paper do not provide a direct link or explicit statement of where this code is available (e.g., a specific repository URL or a mention of its inclusion in supplementary material files).
Open Datasets Yes Our results are obtained by training a deep MLP with width 1024 on the CIFAR-10 dataset
Dataset Splits No The paper mentions training on the CIFAR-10 dataset but does not specify the training, validation, and test splits (e.g., percentages or exact counts). While CIFAR-10 has standard splits, these are not explicitly stated within the paper.
Hardware Specification Yes We performed our experiments on a single NVIDIA A6000 GPU.
Software Dependencies No Our implementation is based on PyTorch. However, no specific version number for PyTorch or any other software dependencies is provided.
Experiment Setup Yes We trained a deep MLP with width 1024 on the CIFAR-10 dataset... using an Adam optimizer... with a learning rate of 1e-3, batch size of 256, and trained for 100 epochs.