In optional coffee lab, the high output value in layer 1 corresponds to bad roast region.

But I think this output must represent the probability of y=1, as the layer 1 also uses sigmoid activation function. So if output is high then the probability of good roast will also be high.

Kindly correct me if Iam wrong or missing something.

Hi @sandeep_kumar13,

Layer 1 using sigmoid is not sufficient for its outcome to represent the probability of y = 1.

The output layer has this representation because this is what the objective function optimizes for.

The objective function does not enforce any hidden layer to share the same representation.

Cheers,

Raymond

@rmwkwok thank you very much for the clarification.

You are welcome @sandeep_kumar13

Raymond