Artificial Neural Network Learning Algorithms: A Comprehensive Report

This report delves into the realm of Artificial Neural Networks (ANNs), examining various learning algorithms. It categorizes ANN learning algorithms into three primary types: supervised, unsupervised, and reinforcement learning. The supervised learning section explores the Multilayer Perceptron and Feed Forward algorithms, providing MATLAB code snippets and execution results to illustrate their functionality. The unsupervised learning section covers clustering, Hebbian learning, and Radial Basis Function algorithms, also including MATLAB examples to demonstrate their implementation and performance. Finally, the report touches upon reinforcement learning algorithms and discusses associated challenges. The report also includes a discussion on recall and precision algorithms. Overall, the report provides a comprehensive overview of ANN learning algorithms, supported by practical examples and code executions, making it a valuable resource for understanding the fundamentals of artificial intelligence.

artificial intelligence

Table of Contents
1. Artificial Neural Network................................................................................................2
2. Different Types of ANN Learning Algorithm................................................................2
3. Supervised learning..........................................................................................................2
3.1 Multilayer perceptron..................................................................................................2
3.2 Feed Forward Learning Algorithm..............................................................................4
4. Unsupervised learning......................................................................................................5
4.1 Clustering algorithm....................................................................................................5
4.2 Hebbian learning algorithm.........................................................................................7
4.3 Radial basis function algorithm...................................................................................8
5. Reinforcement Learning Algorithm..............................................................................10
5.1 Challenges.................................................................................................................10
6. Recall and precision algorithm......................................................................................10
6.1 Mat lab execution......................................................................................................10

1. Artificial Neural Network (ANN)
The neural networks are computing systems inspired by the biological neural
networks. An ANN is based on the collection of connected units which are called as artificial
neurons. These neurons are typically organised in layers. An ANN learning algorithm is
classified into three main types, supervised learning, unsupervised learning and reinforcement
learning algorithms (Berlatsky, 2011).
2. Different Types of ANN Learning Algorithm
An ANN systems is classified into three main categories of neural network
algorithms.
 Supervised learning
 Unsupervised learning
 Reinforcement learning
Part-A
3. Supervised learning
Supervised learning algorithm is divided into various types of learning algorithms.
The most commonly used learning algorithms are listed below
 Multilayer perceptron algorithm
 Feed forward algorithm
3.1 Multilayer perceptron
Multilayer perceptron is a commonly used learning algorithm for mat lab software.
The example for perceptron for the neural network systems is explained below:
Sample code for perceptron
x = [0 0 1 1; 0 1 0 1];
t = [0 1 1 1];
net = perceptron;
net = train(net,x,t);
view(net)
y = net(x);

Figure 1 neural network design for perceptron
Execution program for perceptron
Figure 2program execution for perceptron
Mat lab Execution for the performance of perceptron
Figure 3performance of perceptron

3.2 Feed Forward Learning Algorithm
Feed forward learning algorithm is a two or more layer feed forward network
algorithm which can be used to implement any finite input-output function arbitrarily well
with given number of enough neurons.
Sample code for feed forward algorithm
[x,t] = simplefit_dataset;
net = feedforwardnet(10);
net = train(net,x,t);
view(net)
y = net(x);
perf = perform(net,t,y)
Figure 4 neural network design for feed forward
Program execution for feed forward algorithm
Figure 5 program execution for feed forward

Mat lab execution for performance of feed forward algorithm
Figure 6 performance of feed forward
Part-B
4. Unsupervised learning
Unsupervised learning algorithm is a type of learning algorithms in neural networks.
Unsupervised learning algorithm is a described from a hidden layer of unlabeled system
(Gilat, 2016). The unsupervised learning algorithm is a central case of a problem density
estimation in statistics. The unsupervised learning approaches are mentioned and explained
below
 Clustering algorithm
 Hebbian learning algorithm
4.1 Clustering algorithm
The clustering algorithm uses the kohanen algorithm to create the simple
clustering. The simple clustering process use the following code to create the
clustering.

Sample code for clustering
centerx = [0 0 1 1];
centery = [0 1 0 1];
radius = [0.7 0.6 0.3 0.7];
numSamples = 1000;
x = zeros(2,numSamples);
for i=1:numSamples
j = floor(rand*4)+1;
angle = rand*2*pi;
r = (rand.^0.8)*radius(j);
x(1,i) = centerx(j) + cos(angle)*r;
x(2,i) = centery(j) + sin(angle)*r;
-1 -0.5 0 0.5 1 1.5 2
0
50
100
150
200
250
300
350
Figure 7 plot for simple cluster
Sample for Simple class dataset
load simpleclass_dataset
inputs = simpleclassInputs;
targets = simpleclassTargets;

Program execution for simple cluster
Figure 8 program execution
4.2 Hebbian learning algorithm
The hebbian learning algorithm is an unsupervised learning algorithm. This is being
evolved to make the formation of potentiation. This includes two types of primary visual
cortex known as simple cells and complex cells.
This hebbian learning algorithm is plotted by x versus y followed below:
1 2 3 4 5 6 7 8 9 10
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
Figure 9 Hebbian plot

Mat lab execution for unsupervised Hebb
4.3 Radial basis function algorithm
The sample code for radial basis function algorithm is mentioned below:
Sample code for RBF
x = -3:.1:3;
a = radbas(x);
plot(x,a)
title('Radial Basis Transfer Function');
xlabel('Input p');
ylabel('Output a');

-3 -2 -1 0 1 2 3
Input p
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Output a
Radial Basis Transfer Function
Figure 10 Radial Basis Function
Program execution for RBF
Figure 11 program execution for RBF

Part-C
5. Reinforcement Learning Algorithm
Reinforcement algorithm is a machine learning algorithm or algorithm of the artificial
intelligence. This algorithm behaves within a specific context, in order to maximize its
performance.
5.1 Challenges
Reinforcement algorithm has made many challenges in current learning research. The
main challenge, is too expensive in memory to store the value. This algorithm uses the
techniques which are used for making decision tree or neural network.
6. Recall and precision algorithm
The precision and recall algorithm is used for the recognizing the performance of the
network.
6.1 Mat lab execution code
recall = mean(diag(5)'./[10 10 10 10 10 10])
recall = 0.5000
precision = mean(diag(10)'./sum(10))
precision = 1

References
Berlatsky, N. (2011). Artificial intelligence. Detroit: Green havens Press.
Gilat, A. (2016). MATLAB. New York: Wiley.