Resources
Authors & Affiliations
A. Filipe, Il Park
Abstract
Recent studies have demonstrated that brain areas involved in motor control display common representations of latent trajectories for both stereotypical and generalizable dynamical cortical behavior. However, current research often focuses on analyzing single tasks and individual sessions, missing the opportunity to exploit the structural patterns across datasets. By fitting a unified model to a collection of neural data with shared features, we can increase data usage efficiency and enhance our understanding of common neural computation structures.
Despite these insights, the neuroscience field lacks a comprehensive benchmark dataset to standardize and facilitate multi-task neural analysis, lagging behind fields like computer vision, which benefit from large-scale benchmark datasets. Existing datasets, though relatively large and complex, often lack standardized file formats and benchmarking practices, complicating integration and comparison.
To address these issues, we introduce NeuroTask, a benchmark dataset designed to serve as an intermediate representation between metadata-rich, heterogeneous NWB/MATLAB files and machine learning algorithms. NeuroTask enables unified integration across multiple datasets, promoting extensibility. It includes six electrophysiology datasets from motor and sensory cortical regions, covering seven behavioral tasks across 17 subjects, and incorporates behavior covariates suitable for forecasting and decoding models. The dataset schema allows for easy filtering by subject, session, and task. NeuroTask also provides an API for streamlined data loading and preprocessing.
More than just a dataset, NeuroTask is releasing a couple of benchmarks specifically designed for decoding and forecasting neural data and behavior, as well as for multi-session, multi-animal, and multi-task analyses. By establishing a standardized framework for evaluating multi-task neural population models, NeuroTask promotes systematic, reproducible, and transparent research in neuroscience. We have released the datasets and baseline implementations at : $\href{https://github.com/catniplab/NeuroTask/tree/nwb}{https://github.com/catniplab/NeuroTask/tree/nwb}$. Additionally, we are preparing to launch pretrained multitask models soon, further expanding the scope of this project.