dendritic-spiking-neuron

Author: AllenYolk

A new dendritic computing package dendsn based on PyTorch and SpikingJelly. Draw inspiration from the nonlinear nature of biological dendritic neurons!

Table of Contents:

• Installation
• Modules in dendsn
• TODOs

Installation

Install from source code

From Github

git clone https://github.com/AllenYolk/dendritic-spiking-neuron.git
cd dendritic-spiking-neuron
pip install .

Modules in dendsn

Dendritic Neuron Models

Dendritic neurons are built in a bottom-up manner in this package, and each component is implemented in a separate python script:

• dendritic neuron: model/neuron.py
  • dendrite: model/dendrite.py
    • dendritic compartments: model/dend_compartment.py
    • wiring of the compartments: model/wiring.py
  • soma: model/soma.py
• synapse: model/synapse.py
  • synaptic connection and weights: model/synapse_conn.py
    • Both linear and conv layers are available!
  • synaptic filter: model/synapse_filter.py

The basic assumption is: all the dendritic neurons in the same layer share exactly the same morphology!

Learning Rules

A series of plasticity rules are available in dendsn.learning, whose implementation is based on "monitors" in spikingjelly.

• STDP: learning/stdp.py
• Semi-STDP: learning/semi_stdp.py
  • A simplified form of STDP: trace_post and delta_w_post_pre will be neglected.
• Dendritic Prediction Plasticity: learning/dpp.py
  • See (Urbanczik & Senn, 2014).

Now, these learning rules can only support fully connected weight layers.

Other Modules:

• useful functions: functional.py
• stochastic spiking autograd functions: stochastic_firing.py

TODOs

• Add docstrings and comments to dendsn.learning.
• Extend plasticity rules in dendsn.learning to convolutional layers.

References

• Urbanczik, R., & Senn, W. (2014). Learning by the Dendritic Prediction of Somatic Spiking. Neuron, 81(3), 521–528. https://doi.org/10.1016/j.neuron.2013.11.030