Reverse engineering recurrent neural networks with Jacobian switching linear dynamical systems

Authors: Jimmy Smith, Scott Linderman, David Sussillo

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

Reproducibility Variable Result LLM Response
Research Type Experimental We validate the utility of the model on two synthetic tasks relevant to previous work reverse engineering RNNs. We then show that our model can be used as a drop-in in more complex architectures, such as LFADS, and apply this LFADS hybrid to analyze single-trial spiking activity from the motor system of a non-human primate.
Researcher Affiliation Academia Jimmy T.H. Smith Institute for Computational and Mathematical Engineering Stanford University Stanford, CA 94305 jsmith14@stanford.edu Scott W. Linderman Department of Statistics Stanford University Stanford, CA 94305 scott.linderman@stanford.edu David Sussillo Department of Electrical Engineering Stanford University Stanford, CA 94305 sussillo@stanford.edu
Pseudocode No The paper includes mathematical equations and descriptions of the model and training procedure but does not present any structured pseudocode or algorithm blocks.
Open Source Code Yes Our implementation for the synthetic tasks is available at https://github.com/jimmysmith1919/JSLDS_public.
Open Datasets Yes We used the monkey J single-trial maze data from Churchland et al. [51] using the same setup as Pandarinath et al. [5] to train the LFADS-JSLDS model.
Dataset Splits No The paper mentions evaluating on "held-out trials" which typically refers to a test set. It does not explicitly specify the proportions or sizes of training, validation, and test splits for all experiments. For the monkey data, it refers to using the "same setup as Pandarinath et al. [5]" without detailing the splits in this paper.
Hardware Specification No The paper does not specify any hardware details such as GPU models, CPU types, or cloud computing resources used for running the experiments.
Software Dependencies No The paper mentions using GRUs and a TensorFlow toolbox (via a citation to [32]) but does not list specific version numbers for any software libraries, frameworks, or programming languages used in the implementation.
Experiment Setup Yes The RNNs used in this experiment were GRUs [48] with a state dimension of 100 and a linear readout function. See Section A.3 of the Appendix for additional experiment details. ... We used a vanilla RNN with a state dimension of 128 for the co-trained RNN and a linear readout function. ... In practice, we have found these hyperparameters straightforward to select (see Section A.1 in the Appendix for a more detailed discussion).