Scribd
Upload
2 views

quiz1 solution

The document describes a neural network using negative sigmoid units and a specific error function to calculate weight updates after one epoch, resulting in new weights for w2, w3, and w1. Additionally, it applies the gradient descent method to a polynomial function, tracking the trajectory of x starting from an initial value. The results indicate the updated weights and the iterative process of optimization for the given function.

Uploaded by

bhavdeep3singh
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
2 views

quiz1 solution

The document describes a neural network using negative sigmoid units and a specific error function to calculate weight updates after one epoch, resulting in new weights for w2, w3, and w1. Additionally, it applies the gradient descent method to a polynomial function, tracking the trajectory of x starting from an initial value. The results indicate the updated weights and the iterative process of optimization for the given function.

Uploaded by

bhavdeep3singh
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

1.

In the neural network shown in Figure, all the neurons use negative sigmoid units meaning that
1
their output is computed as o = − . Additionally, the network uses a non-standard error
1 + e −net
function E =
1
(2t − 2o )2 . If the target outputs are and and the learning rate is
2
0.2, determine all the weights after one epoch.

w2=3;w3=2;x=2; 3 y2
4
x=2
2
y3
=-0.99966 net2=w2*a1; net3=w3*a1;
o1 = -0.04747 o2 = -0.11927
δ2=4*(t1-o1)*o1*(1+o1); δ2=del of first output neuron = -0.3703
δ3 =4*(t2-o2)*o2*(1+o2); δ3=del of second output neuron = -0.2602
δ1=a1*(1+a1)*( δ2*w2+ δ3*w3); δ1 del of the hidden neuron = 0.0005469
w2n=w2+0.2* δ2*a1; new w2 = 3.074
w3n=w3+0.2* δ3*a1; new w3 = 2.052
w1n=w1+0.2* δ1*x ; new w1 = 4.0002

2. Apply gradient descent method to 2x12+2x1x2+21x22 and find the value of x after one
iteration. The trajectory of x with a starting value x(0) = [-2 1]T is shown in Figure 1
-0.904999345720556 -0.900013085588878 -1
-1.714998037161668 -0.805012431309434 -2
-4.144994111485005 -0.520010468471102 -5
-4.954992802926117 -0.425009814191659 -6
-5.764991494367229 -0.330009159912215 -7

You might also like