STNDT: Modeling Neural Population Activity with Spatiotemporal Transformers

Authors: Trung Le, Eli Shlizerman

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

Reproducibility Variable Result LLM Response
Research Type Experimental We show that our model achieves state-of-the-art performance on ensemble level in estimating neural activities across four neural datasets, demonstrating its capability to capture autonomous and non-autonomous dynamics spanning different cortical regions while being completely agnostic to the specific behaviors at hand.
Researcher Affiliation Academia Trung Le University of Washington Seattle, WA tle45@uw.edu Eli Shlizerman University of Washington Seattle, WA shlizee@uw.edu
Pseudocode No The provided text does not contain any structured pseudocode or algorithm blocks.
Open Source Code Yes Code is available at https://github.com/shlizee/STNDT
Open Datasets Yes We evaluate our model performance on four neural datasets in the publicly available Neural Latents Benchmark [23]: MC_Maze, MC_RTT, Area2_Bump, and DMFC_RSG.
Dataset Splits Yes Bayesian hyperparameter tuning: We follow [47] to use Bayesian optimization for hyperparameters tuning. We observe that the primary metrics co-bps are not well correlated with the mask loss (see Figure 1 in the Appendix , while co-bps, vel R2, psth R2 and fp-bps are more pairwise correlated. Therefore, we run Bayesian optimization to optimize co-bps for M models then select the best N models as ranked by validation co-bps, and ensemble them by taking the mean of the predicted rates of these N models.
Hardware Specification No The main text of the paper does not specify the hardware used for experiments. It defers to the Appendix, which is not provided in the given text.
Software Dependencies No The provided text does not specify any software dependencies with version numbers.
Experiment Setup Yes Bayesian hyperparameter tuning: We follow [47] to use Bayesian optimization for hyperparameters tuning. ... We used 2 heads for all reported models. ... Did you specify all the training details (e.g., data splits, hyperparameters, how they were chosen)? [Yes] Please see Appendix.