I'm using Alexnet and want to fuse two signals to observe the performance compared to individual training. I see that it is possible to do that by training these two inputs seperately and fuse the last activation maps before the first fully connected layer. So the network combines both features and classifies like a single architecture. However I don't know how to concetenate two layers to form a single one and classify for AlexNet or any other pre-trained network. Do you know how to do that? Thanks.