Problem Description
Implement (in C++) a fully connected feed-forward neural network that con-sists of 3 input neurons, 2 hidden layer neurons and 1 output neuron. The hidden layer neurons and output neuron use the Sigmoid (chapter 4 [Mitchell, 1997]) activation function.
The ANN input nodes 1, 2 and 3, respectively, have the following inputs:
x = [ 1:30, 2:70, 0:80 ]
The ANNs expected output is:
y = 0:36
Table 1 speci es the weights connecting the inputs to the hidden layer neu-rons. Note that each column denotes a given node connected to a given hidden layer. For example, the top value in column 1 is the value of the weight con-necting input node 1 to hidden layer node 1, and the bottom value in column 1 is the value of weight connecting input node 1 to hidden layer node 2.
The bias values for the hidden layer neurons 1 and 2, respectively, are:
b = [ 0:1, 0:3 ]
Table 1: Values of weights connecting input to hidden layer nodes.
Input 1
Input 2
Input 3
0:1
0:2
0:5
0:4
1:0
0:6
Question 1:
Given the ANN input x, what are the output values of hidden layer neurons 1 and 2 ?
Question 2:
Given that the weights from hidden layer nodes 1 and 2, respectively, are:
w = [ 0:8, 1:0 ]
And the bias value for the output node is: b = 0:3, and the ANN input x, what is the output value of the output neuron ?
Question 3:
Given this output, what is the Mean Squared Error for ANN input x.
In a ZIP le, place the source code, executable, and a text le containing your list of training examples, as well as answers to questions 1, 2 and 3. Upload the ZIP le to Vula before 10.00 AM, 7 September, 2018.
References
[Mitchell, 1997] Mitchell, T. (1997). Machine Learning. McGraw Hill, New York, USA.
2
