Signing the Supermask: Keep, Hide, Invert

Authors: Nils Koster, Oliver Grothe, Achim Rettinger

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

Reproducibility Variable Result LLM Response
Research Type Experimental The following paragraphs present the experiments conducted in order to assess the validity and effectiveness of fixed threshold signed Supermasks. The code for the experiments can be found here.
Researcher Affiliation Academia Nils Koster Department of Applied Econometrics Karlsruhe Institute of Technology nils.koster@kit.edu Oliver Grothe Department of Analytics and Statistics Karlsruhe Institute of Technology oliver.grothe@kit.edu Achim Rettinger Department of Computational Linguistics and Digital Humanities Trier University rettinger@uni-trier.de
Pseudocode No The paper includes mathematical formulations and descriptions of processes but does not present any explicitly labeled pseudocode blocks or algorithms.
Open Source Code Yes The code is available here.
Open Datasets Yes The fully-connected (FCN) model is trained on MNIST (Le Cun et al., 2010) and all CNNs plus Res Net20 on CIFAR-10 (Krizhevsky, 2009). Res Net56 and Res Net110 are trained on CIFAR-100 (Krizhevsky, 2009)... All datasets used in this work are provided by tensorflow here.
Dataset Splits No The paper specifies datasets used for training and testing, and describes preprocessing steps like image standardization and augmentation (for CIFAR-100), but it does not provide explicit details about validation set splits (e.g., percentages or sample counts for a validation set).
Hardware Specification No The paper states that 'all experiments were run on a single GPU' but does not specify the model or type of GPU, CPU, or any other detailed hardware specifications.
Software Dependencies No The paper mentions that 'The code is written in Python, mainly using tensorflow and numpy.' However, it does not provide specific version numbers for Python, TensorFlow, NumPy, or any other software libraries.
Experiment Setup Yes Tables 8, 9, 10, and 11 in Appendix B summarize hyperparameter choices for training both baseline and signed Supermask models, including learning rates, decay rates, weight decay, momentum, and iterations.