Fall EEL-5840 2009
Homework Dr. Arroyo
EEL5840: Machine Intelligence I: Homework #4
(Due Tuesday September 22, 2009)
Problem 2 Consider a one-hidden layer feedforward neural network, fully connected between layers, with 2 inputs, 2 hidden units in the first layer, and 1 output. Assume sigmoidal activation functions . a. How many total independent weights are contained in this neural network? Show Work! No Work, No Credit. Answer: ____ independent weights.
b. Without any local variables, how many additions, multiplications, and function evaluations ( , ') are required to compute the output of the neural network in terms of the inputs and weights the of the neural network? Show Work! No Work, No Credit.
Additions:___________
Multiplications:___________
Function Evaluations:__________
c. Using the backpropagation algorithm, how much total computation is required [additions, multiplications and function evaluations ( , ')] to compute the error derivatives with respect to all the weights in the neural network, given a single training pattern <x1,x2,y>and the error measure E= (y-z)2? Show Work! No Work, No Credit. Forward Pass: Additions:___________ Multiplications:___________ Function Evaluations:__________
Backward Pass:
EEL-5840 Fall 2009
Homework Dr. Arroyo