Download
SpikingJelly is an open-source deep learning framework for spiking neural networks that is primarily built on top of PyTorch and aimed at neuromorphic computing research. The project provides the components needed to build, train, and evaluate neural models that communicate through discrete spikes rather than the continuous activations used in conventional artificial neural networks. This makes it especially relevant for researchers interested in biologically inspired computing, event-driven processing, and energy-efficient AI systems. The framework includes neuron models, surrogate gradient training methods, encoding strategies, network components, and utilities for simulation and experimentation, allowing users to develop a wide variety of spiking architectures. It also supports integration with familiar PyTorch workflows, which lowers the barrier for machine learning practitioners who want to explore spiking approaches without abandoning mainstream tooling.
Features
- PyTorch-based framework for building spiking neural networks
- Support for neuron models, encoding methods, and temporal dynamics
- Surrogate gradient training for end-to-end optimization
- Tools for neuromorphic computing and event-driven AI research
- Suitable for experimentation with vision, time-series, and spike-based tasks
- Designed for both research prototyping and educational use
Project Samples
