Local plasticity rules can learn deep representations using self-supervised contrastive predictions

Authors: Bernd Illing, Jean Ventura, Guillaume Bellec, Wulfram Gerstner

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

Reproducibility Variable Result LLM Response
Research Type Experimental 4 Empirical results
Researcher Affiliation Academia Bernd Illing Jean Ventura Guillaume Bellec Wulfram Gerstner {firstname.lastname}@epfl.ch Department of Computer Science & Department of Life Sciences École Polytechnique Fédérale de Lausanne 1015 Switzerland
Pseudocode No The paper describes the CLAPP algorithm but does not provide it in a formal pseudocode block or a clearly labeled algorithm section.
Open Source Code Yes 2Our code is available at https://github.com/EPFL-LCN/pub-illing2021-neurips
Open Datasets Yes We first consider the STL-10 image dataset [Coates et al., 2011].
Dataset Splits No The paper mentions training on the unlabelled part of the STL-10 dataset and evaluating on the test set, but it does not explicitly provide the specific percentages or counts for training, validation, or test splits for the main CLAPP training or the linear classifier training.
Hardware Specification Yes We use 4 GPUs (NVIDIA Tesla V100-SXM2 32 GB) for data-parallel training, resulting in a simulation time of around 4 days per run.
Software Dependencies No The paper mentions "Pytorch" in a citation, but it does not specify the version numbers for PyTorch or any other software dependencies used in their experiments.
Experiment Setup Yes Other hyper-parameters and data-augmentation are taken from Löwe et al. [2019], see Appendix B. We then train a 6-layer VGG-like [Simonyan and Zisserman, 2015] encoder (VGG-6) using the CLAPP rule (Equations 6 8). Training is performed on the unlabelled part of the STL-10 dataset for 300 epochs.