Project 1 Assignment This assignment involves using backpropagation (BP) learnin
Place your order now for a similar assignment and have exceptional work written by our team of experts, At affordable rates
For This or a Similar Paper Click To Order Now
Project 1 Assignment This assignment involves using
backpropagation (BP) learning procedure to train a multi-layer feedforward
network to compute the 4-bit parity function. The bipolar representation is
recommended. There are total of sixteen training samples (x:class(x)) where
vector x is a unique 4-bit input pattern
and class(x) is its corresponding target function
value. It is well known that class(x)=1 if x contains
odd number of 1 ‘s, and class(x)=−1, otherwise. You are
asked to implement two versions of BP learning, one using the standard BP
procedure and the other with a momentum term, and compare their respective
learning performance. Besides the source code of your program, you should
submit a written report of the following: 1. The architecture of your network
(number of hidden layers and their sizes, the sigmoid function you choose,
etc.) 2. The parameters you selected for the network, such as the learning
rate, the initial weight matrix. 3. Criteria used for termination (e.g.,
acceptable error). 4. Whether learning is conducted in sequential or batch
mode. 5. For each of the two versions, report 1) The number of iterations (or
the number of epochs if you choose batch mode) the learning takes to its
termination. 2) The learned weights (including weights for the dummy/bias
nodes). 3) The output value of the output node when each of the sixteen input
patterns is applied to the learned network. 6. Any other issues you wish to
report and discuss. To highlight the linear inseparability nature of this
problem, you are also asked to apply Adaline learning procedure to the same
parity function. After the training is terminated, report the decision boundary
(i.e., the linear function) found by the learning process, and which of the
sixteen inputs are correctly classified and which are not.
