Question about the tfc-w2a2 model.

[liuyh-Horizon]

Hi, everyone. Sorry to disturb you. I just have one very beginner question about the example network model of FINN.

I'm trying to understand the model structures. In the "tfc-w2a2" model, I found that before the final output of the top-k block, there are 4 blocks, which looks like a batch normalization computation.

However, the question is: if we assume the output value of MatMul is x={x1, x2, ..., xn}, after the BN computing, the input of the top-k block will be in(x)=[(x-a)/b]*c+d. which can be converted as in(x)=(c/b)x+(d-ac/b). This is a linear function of x. Therefore, if we want to find the biggest value in (x), actually it will be as same as finding the biggest value in x. This BN computing looks like it can be ignored.

Hence, If we only want to know the index of the biggest value, it's possible to ignore these four blocks before the top-k block?

Thank you so much for your help.