How does a hidden layer work in a feed forward net?

q1. on the first layer of the hidden layer , wont all the neurons look the same after processing ?
q2. did not understand how 2 sigmoid neurons in layer 1 could magically work for a doughnut shaped structure. did i miss something very basic ?


  1. Each neuron is just an approximation function, and learns different parameters and hence capture different features.
  2. Please rewatch the theory lectures to understand better.