[thomasahle]

> Fully connected neural networks are hierarchies of logistic regression nodes.

Only if you use softmax ss your activation function.

[bjornsing]

You mean sigmoid activation function?

[thomasahle]

If we are talking about "hierarchies of logistic regression nodes" we have to define how to extend logistic regression to multiple outputs. The most common approach is Multinomial logistic regression: https://en.wikipedia.org/wiki/Logistic_regression#Extensions . Other times sigmoid might be the right answer.