Towards robust and generalizable representations of extracellular data using contrastive learning

Authors: Ankit Vishnubhotla, Charlotte Loh, Akash Srivastava, Liam Paninski, Cole Hurwitz

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We validate our method across multiple high-density extracellular recordings. All code used to run CEED can be found at https://github.com/ankitvishnu23/CEED. (Lines 9-11) and we find that CEED outperforms both PCA and the non-linear autoencoder using raw waveforms or denoised waveforms across all three datasets introduced in Section 4. (Lines 285-287)
Researcher Affiliation Collaboration Ankit Vishnubhotla Columbia University New York av3016@columbia.edu Charlotte Loh MIT Massachusetts cloh@mit.edu Liam Paninski Columbia University New York liam@stat.columbia.edu Akash Srivastava MIT-IBM Massachusetts Akash.Srivastava@ibm.com Cole Hurwitz Columbia University New York ch3676@columbia.edu (Lines 1-5)
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code Yes All code used to run CEED can be found at https://github.com/ankitvishnu23/CEED. (Lines 10-11)
Open Datasets Yes To train and evaluate our model, we make use of two publicly available extracellular recordings published by the International Brain Laboratory (IBL): the DY016 and DY009 recordings [54]. (Lines 216-218)
Dataset Splits Yes The first dataset was extracted from the DY016 extracellular recording. It consisted of a 10 unit train and test dataset... For this dataset, we constructed training sets of 200 or 1200 spikes per unit with a test set of 200 spikes per unit. (Lines 223-227)
Hardware Specification No The paper mentions running experiments on 'large-scale, multi-gpu clusters' for the transformer model and a 'single GPU' for the MLP-based architecture, but does not specify exact GPU or CPU models, memory, or other detailed hardware specifications.
Software Dependencies No The paper does not provide specific version numbers for software dependencies or libraries used for implementation.
Experiment Setup Yes The MLP encoder is a straightforward model that consists of three layers with sizes [768, 512, 256] and ReLU activations between them. (Lines 204-206) and For all baselines, we sweep across (3,5,7,9) principal components and 3-11 channel subset sizes. (Lines 263-264)