f3ca0de6ca8c1b6b75ff0d75cc9d0e19adca7648,bindsnet/learning/__init__.py,Hebbian,_conv2d_connection_update,#Hebbian#,402

Before Change

            self.source.x, kernel_height, kernel_width, padding=padding, stride=stride
        )

        target_x = self.target.x.permute(, , , ).view(out_channels, -1)
        source_s = im2col_indices(
            self.source.s.float(),
            kernel_height,
            kernel_width,
            padding=padding,
            stride=stride,
        )
        target_s = self.target.s.permute(, , , ).view(out_channels, -1).float()

        // Pre-synaptic update.
        pre = target_x @ source_s.t()
        self.connection.w += self.nu[0] * pre.view(self.connection.w.size())

        // Post-synaptic update.
        post = target_s @ source_x.t()
        self.connection.w += self.nu[1] * post.view(self.connection.w.size())

        super().update()

After Change

        
        out_channels, _, kernel_height, kernel_width = self.connection.w.size()
        padding, stride = self.connection.padding, self.connection.stride
        batch_size = self.source.batch_size

        // Reshaping spike traces and spike occurrences.
        source_x = im2col_indices(
            self.source.x, kernel_height, kernel_width, padding=padding, stride=stride
        )
        target_x = self.target.x.view(batch_size, out_channels, -1)
        source_s = im2col_indices(
            self.source.s.float(),
            kernel_height,
            kernel_width,
            padding=padding,
            stride=stride,
        )
        target_s = self.target.s.view(batch_size, out_channels, -1).float()

        // Pre-synaptic update.
        pre = torch.bmm(target_x, source_s).sum(dim=0)
        self.connection.w += self.nu[0] * pre.view(self.connection.w.size())

        // Post-synaptic update.
        post = torch.bmm(target_s, source_x).sum(dim=0)
        self.connection.w += self.nu[1] * post.view(self.connection.w.size())

        super().update()

Instances