Reconstruction a neuron synapse

1

Dear all,
I want to reconstruct a synapse in the first layer of full-connection, which shows in Fig. Could you help me.

RECONS_NN
RECONS_NN1280×720 43.9 KB

2

If I understand your use case correctly, you would like to transform the left layer to the right one.
In that case I assume the left layer is defined as nn.Linear(784, 1, bias=False) and you could replace it with:

class MyModel(nn.Module):
    def __init__(self):
        super(MyModel, self).__init__()
        self.fc1a = nn.Linear(392, 1, bias=False)
        self.fc1b = nn.Linear(392, 1, bias=False)
        self.fc2 = nn.Linear(2, 1, bias=False)
        
    def forward(self, xa, xb):
        xa = self.fc1a(xa)
        xb = self.fc1b(xb)
        x = torch.cat((xa, xb), dim=1)
        x = self.fc2(x)
        return x

model = MyModel()
batch_size = 2
xa = torch.randn(batch_size, 392)
xb = torch.randn(batch_size, 392)
out = model(xa, xb)

Note that I haven’t added any activation functions etc. as it’s unclear from the figure, if this is desired.

1 Like