Neural networks

Question 1

what is the principle of backpropagation

Answer 1

The idea is to propagate the signal pf the hidden layer , where the error is computed at each output allowing to adjust the weight accordingly

Question 2

what do we use to minimize the criterion for regression

Answer 2

Generally, we either use gradient descent or ascent to allow the minimization

Question 3

what are the main component of a neural network

Answer 3

intput layer hidden layer(s) output layer

Question 4

what is the global purpose of NN

Answer 4

all the connected weights are adapted to optimize a supervised criterion which is an iterative optimization based on the gradient ascent/ descent

Question 5

what is an error criterion

Answer 5

A non linear function of the weights with numerous local minima

Question 6

what is the difference between global and local approach

Answer 6

for Gloabl apprach which correspond to supevised problems it means that all the connections are adapted to optimize a supervised criterion and the optimization is based on criterion

for local approach unsupervised problem meaning adaptation of reduced number of weights