A PyTorch library for Spiking Dendrite-Enabled neurons. This library implements a hardware-aware Dendrite within PyTorch, for use with the SNNTorch library or as a stand-alone PyTorch module.
TorchDendrite is a module that implements a hardware-inspired dendritic cable function using Pytorch and SNNtorch. The module contains trainable parameters and is compatible with both spiking and non-spiking neural networks.
This project uses Hatch to manage the package and project. To install, make sure you are running Python >= 3.10.
Once you have cloned the repository, you can install it by navigating to the project directory and running:
hatch installTo get started with using TorchDendrite, you can refer to the examples and Jupyter notebooks provided in the notebooks directory. These notebooks demonstrate how to use TorchDendrite to add dendrite layers to a PyTorch network and how to use it with SNNtorch.
When training models with dendrites, it is recommended to use the Adam optimizer as it tends to perform better with dendritic layers.
torchdendrite is distributed under the terms of the MIT license.
Currently in development; Examples and documentation are WIP.