Demystifying and Generalizing BinaryConnect

Authors: Tim Dockhorn, Yaoliang Yu, Eyyüb Sari, Mahdi Zolnouri, Vahid Partovi Nia

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct experiments on CIFAR-10 and Image Net, and verify that PC achieves competitive performance.
Researcher Affiliation Collaboration Tim Dockhorn University of Waterloo Yaoliang Yu University of Waterloo Eyyüb Sari Huawei Noah s Ark Lab Mahdi Zolnouri Huawei Noah s Ark Lab Vahid Partovi Nia Huawei Noah s Ark Lab
Pseudocode No The paper describes algorithms using mathematical notation and textual descriptions but does not include formal pseudocode blocks or algorithm listings with specific labels like 'Algorithm 1'.
Open Source Code No We will provide the code once it has been approved by our internal process.
Open Datasets Yes We conduct experiments on CIFAR-10 [24] using Res Net20 and Res Net56 [19] (...) We perform a small study on Image Net [12] using Res Net18 [19].
Dataset Splits No The paper mentions using CIFAR-10 and ImageNet but does not explicitly detail the training/validation/test dataset splits (e.g., percentages or sample counts) or how a validation set was used, beyond mentioning test accuracy.
Hardware Specification Yes We conduct experiments on a single NVIDIA Quadro P6000 GPU.
Software Dependencies No The paper mentions PyTorch but does not provide specific version numbers for any software dependencies.
Experiment Setup Yes we increase the parameter ρ (or equivalently ϱ) linearly: ρt = (1 + t/B) ρ0. In contrast to Bai et al. [5], however, we increase ρ after every gradient step rather than after every epoch as this is more in line with our analysis. We treat ρ0 as a hyperparameter for which we conduct a small grid search.