In fact, a neural net is “just” a stack of logistic regressions if you only use sigmoid activations.