Description
Use your own implementation of the feed-forward neural network to
a. Solve the (modified) XOR classification problem, as shown below, using
an architecture with one hidden layer composed of 2 units.
Plot the loss function value throughout the training process. Once
convergence is achieved, plot the resulting decision surface together
with the input data.
𝐶𝐶1 = ��
−1
−1
�,�
1
1
��; 𝐶𝐶2 = ��
−1
1 �,� 1
−1
��
b. Solve the regression problem described below.
i. Generate the following dataset. For those programming in Python,
the dataset is provided in the attached Excel file.
ii. Network architecture: One hidden layer composed of 3 hyperbolic
tangent units
iii. Repeat with 20 hyperbolic tangent units
Plot the loss function value throughout the training process. Once
convergence is achieved, plot the resulting model together with the
input data and report the training error on the figure.
When executed, your code must display all required plots and data at once.
Upload your .m or .py file to Blackboard prior to the deadline. Name your files as
Lastname_Firstname_Project5.*
rng(100)
X=2*rand(1,50)-1;
T=sin(2*pi*X)+0.3*randn(1,50);
