This module covers containers that can encapsulate SNN modules. Through these containers, SNN modules based on matterhorn_pytorch.snn.Module can form spatio-temporal computing networks.
Containing matterhorn_pytorch.snn modules: Spatial, Temporal, and Sequential.
You can read the documentation to learn more about the usage of classes and variables.
This module contains decoders that transform spike sequences into analog values by statistically analyzing information along the temporal dimension. It serves as a bridge for the output of SNNs to be used as input for ANNs.
Containing matterhorn_pytorch.snn modules: SumSpikeDecoder, AvgSpikeDecoder, MinTimeDecoder, and AvgTimeDecoder.
You can read the documentation to learn more about the usage of classes and variables.
This module contains encoders that encode analog values along the temporal dimension into spike sequences. It serves as a bridge for the output of ANNs to be used as input for SNNs.
Containing matterhorn_pytorch.snn modules: DirectEncoder, PoissonEncoder, and TemporalEncoder.
You can read the documentation to learn more about the usage of classes and variables.
This module defines some common functions in SNNs, such as the Heaviside step function, etc.
You can read the documentation to learn more about the usage of functions.
The units in this module, both input and output are spike sequences. It contains flattening layers, pooling layers, etc.
Containing matterhorn_pytorch.snn modules: STDPLinear, MaxPool1d, MaxPool2d, MaxPool3d, AvgPool1d, AvgPool2d, AvgPool3d, Flatten, Unflatten, Dropout, Dropout1d, Dropout2d, and Dropout3d.
You can read the documentation to learn more about the usage of classes and variables.
A base class inherited by all modules in matterhorn_pytorch.snn, serving as the skeleton of matterhorn_pytorch.snn.
Containing matterhorn_pytorch.snn modules: Module.
You can read the documentation to learn more about the usage of classes and variables.
This module defines the soma of SNNs, where the input is the membrane potential (analog value), and the output is a spike sequence. It contains IF, LIF neurons, etc.
Containing matterhorn_pytorch.snn modules: IF, LIF, QIF, ExpIF, Izhikevich, KLIF, and LIAF.
You can read the documentation to learn more about the usage of classes and variables.
matterhorn_pytorch.snn.firing: Surrogate Gradients for the Heaviside Step Function of Spiking Neurons
This module defines surrogate gradients for the Heaviside step function of spiking neurons. The specific definition can be found in reference [1].
Containing matterhorn_pytorch.snn modules: Rectangular, Polynomial, Sigmoid, and Gaussian.
You can read the documentation to learn more about the usage of classes and variables.
This module defines the synapse of SNNs, where the input is a spike sequence and the output is the membrane potential (analog value). It contains operations such as fully connected layers, convolutions, etc.
Containing matterhorn_pytorch.snn modules: Linear, Conv1d, Conv2d, Conv3d, ConvTranspose1d, ConvTranspose2d, ConvTranspose3d, BatchNorm1d, BatchNorm2d, BatchNorm3d, and LayerNorm.
You can read the documentation to learn more about the usage of classes and variables.
[1] Wu Y, Deng L, Li G, et al. Spatio-temporal backpropagation for training high-performance spiking neural networks[J]. Frontiers in neuroscience, 2018, 12: 331.