Overview
NeuroTorch is a neuroscience-oriented machine learning library built on PyTorch. It provides a flexible and intuitive framework to design and train models that incorporate biologically inspired neural dynamics such as spiking neural networks (SNNs) and Wilson–Cowan models. By offering a full training pipeline with Backpropagation Through Time (BPTT) and Truncated BPTT, NeuroTorch makes it easy to experiment with time-dependent neural activity and biologically realistic architectures.
The project aims to bridge the gap between machine learning and computational neuroscience, giving neuroscientists user-friendly ML tools and enabling ML researchers to explore biologically plausible learning algorithms. It serves as a common platform where both communities can build, train, and evaluate spiking or rate-based models using standardized interfaces.
NeuroTorch was developed as part of a postgraduate research project at Université Laval, led by Jérémie Gince supervised by Simon V. Hardy and Patrick Desrosiers.
