Is there a way to element-wise multiply the outputs of two layers that have different dimensions using broadcasting? For example, layer A in my network outputs 5-vectors and layer B outputs 51x51x5-tensors and I'd like to multiply the output of A by each of the 51^2 5-vectors the make up the output of B. Thanks in advance!
It should be easy to write such a layer in CPU backend and just use Julia's broadcasting feature. However, for GPU, a bit work is required. There is no existing layer that could do this.
Is there a way to element-wise multiply the outputs of two layers that have different dimensions using broadcasting? For example, layer A in my network outputs 5-vectors and layer B outputs 51x51x5-tensors and I'd like to multiply the output of A by each of the 51^2 5-vectors the make up the output of B. Thanks in advance!