1_functional.md

matterhorn_pytorch.training.functional

Back to matterhorn_pytorch.training

English

中文

Module Introduction

This module contains functions for custom training in the matterhorn_pytorch.training module.

matterhorn_pytorch.training.functional.stdp

Function for Spike Timing-Dependent Plasticity (STDP). It computes the weight update based on input spike train and output spike train.

stdp(
    delta_weight: torch.Tensor,
    input_spike_train: torch.Tensor,
    output_spike_train: torch.Tensor,
    a_pos: float,
    tau_pos: float,
    a_neg: float,
    tau_neg: float
) -> torch.Tensor

Parameters

delta_weight (torch.Tensor): Weight increment matrix $\Delta w$, with size [O, I].

input_spike_train (torch.Tensor): Input spike train of the layer, with size [T, B, I].

output_spike_train (torch.Tensor): Output spike train of the layer, with size [T, B, O].

a_pos (float): STDP parameter $A_{+}$.

tau_pos (float): STDP parameter $\tau_{+}$.

a_neg (float): STDP parameter $A_{-}$.

tau_neg (float): STDP parameter $\tau_{-}$.

Returns

delta_weight (torch.Tensor): Computed weight increment matrix $\Delta w$, with size [O, I].

Example Usage

import torch
import matterhorn_pytorch as mth
import matterhorn_pytorch.snn.functional as SF
from matterhorn_pytorch.training.functional import stdp


delta_weight = torch.zeros(4, 6, dtype = torch.float) # [O = 4, I = 6]
input_spike_train = SF.to_spike_train(torch.rand(8, 1, 6)) # [T = 8, B = 1, I = 6]
output_spike_train = SF.to_spike_train(torch.rand(8, 1, 4)) # [T = 8, B = 1, O = 4]
delta_weight = stdp(delta_weight, input_spike_train, output_spike_train, 0.2, 2, 0.2, 2)
print(delta_weight)