Neural Report.docx

PROJECT REPORT “ Kohonen Self organizing map

”

Submitted for CAL in B.Tech Neural Networks and Fuzzy Logic (EEE1007) Submitted by

Abhilash Pani (15BEE1030) Akash Singh (15BEE1048) Vaibhav (15BEE1147)

Slot: E1

Name of faculty: DEEPA T. (SCHOOL OF ELECTRICAL ENGINEERING)

April 2017

i

CERTIFICATE This is to certify that the Project work entitled “Kohonen Self organizing map” that is being submitted by “Abhilash Pani, Akash Singh and Vaibhav” for CAL in B.Tech Neural Networks and Fuzzy Logic (EEE1007) is a record of bonafide work done under our supervision. The contents of this Project work have not been submitted for any other CAL course.

Place : Chennai Date : May2nd, 2017

Signature of students: Abhilash Pani

Akash Singh

Vaibhav

Signature of Faculty: DEEPA T.

ii

CONTENTS

ACKNOWLEDGEMENTS………………………………………………………...iv ABSTRACT………………………………………………………………………iv 1. Introduction………………………………………………….......................v 2. Conclusion and applications……..................................................................xi 3. References...................................................................................................xi

ACKNOWLEDGEMENTS ACKNOWLEDGEMENTS

iii

ACKNOWLEDGEMENTS

It is our duty to thank the faculty of SELECT, Dr Angalaeswari, for giving us valuable inputs and most importantly our faculty, Ms Deepa T, for her invaluable guidance and support throughout the entire semester and giving us this opportunity to make this project possible

Abhilash Pani Registration number 15BEE1030 Akash Singh Registration number 15BEE1048 Vaibhav Registration number 15BEE1147

iv

ABSTRACT This project report is a detailed explanation about the idea that went into building the code for Kohonen self-organizing map. The code, built to be used in MATLAB, works to different parts of he network to respond similarly to certain input patterns. It has been programmed in such a way that it can present the result almost immediately. The aim here is to analyze the use of Artificial Neural Network in clinical and diagnostic approaches through selected examples. The use of neural network in diagnosing fever, through definite medical example, showing the output state. Segmentation of Brain Tumor from Magnetic Resonance Images, using Kohonen Model of Artificial Neural Network. MATLAB code for analyzing non-morphological distinction between glomerular and tubular renal disease, using Kohonen Model and obtaining definite state of result.

v

1. Introduction What is a Self organizing map? A self-organizing map (SOM) or self-organizing feature map (SOFM) is a type of artificial neural network (ANN) that is trained using unsupervised learning to produce a low-dimensional (typically twodimensional), discretized representation of the input space of the training samples, called a map, and is therefore a method to do dimensionality reduction. A self-organizing map consists of components called nodes or neurons. Associated with each node are a weight vector of the same dimension as the input data vectors, and a position in the map space. The usual arrangement of nodes is a two-dimensional regular spacing in a hexagonal or rectangular grid. The self-organizing map describes a mapping from a higher-dimensional input space to a lowerdimensional map space. The procedure for placing a vector from data space onto the map is to find the node with the closest (smallest distance metric) weight vector to the data space vector. While it is typical to consider this type of network structure as related to feedforward networks where the nodes are visualized as being attached. The goal of learning in the self-organizing map is to cause different parts of the network to respond similarly to certain input patterns. This is partly motivated by how visual, auditory or other sensory vi

information is handled in separate parts of the cerebral cortex in the human brain. The weights of the neurons are initialized either to small random values or sampled evenly from the subspace spanned by the two largest principal component eigenvectors. With the latter alternative, learning is much faster because the initial weights already give a good approximation of SOM weights. The network must be fed a large number of example vectors that represent, as close as possible, the kinds of vectors expected during mapping. The examples are usually administered several times as iterations. The training utilizes competitive learning. When a training example is fed to the network, its Euclidean distance to all weight vectors is computed. The neuron whose weight vector is most similar to the input is called the best matching unit (BMU). The weights of the BMU and neurons close to it in the SOM lattice are adjusted towards the input vector. The magnitude of the change decreases with time and with distance (within the lattice) from the BMU.

Neural Networks in Medical Sciences Diagnosing Fever: The rescaled temperature for patients are noted and targets are found using threshold. Using Gradient descent method along with Logistic function and delta rule for updation of weights. Random values for weights, and bias are initialized along with learning rate. Output is obtained using logistic activation function. Square error is calculated, weights are updated using Delta rule. The iteration or epoch is done until the square error is zero. Locating the brain tumor using Kohonen model Contrast enhancement, for proper segmentation, the intensity is varied to distinguish healthy parts of brain. Cranium Removal, the bones around the brain mass is removed. Analyzing Renal diseases using Kohonen model by MATLAB. vii

Random input and weights are initialized. Squared Euclidian distance is calculated, winning cluster is found out by comparing for minimum distance. Updating the weight matrix, and iterating for the given number. After the iteration is over displaying the output, using drawnow command for flushing the input every time.

The Code % Self organizing map Kohonen clc clear all close all for i=1:1000 Input_1(i)=rand;%Inputs been given randomly Input_2(i)=rand;%Inputs been given randomly end display(Input_1) display(Input_2) for j1=1:10 for j2=1:10 Weight_1(j1,j2)=rand*0.04+0.6;%weights have been randomly Weight_2(j1,j2)=rand*0.04+0.6;%weights have been randomly end end %display(Weight_1) %display(Weight_2) figure(1) plot(Input_1,Input_2,'.b') hold on plot(Weight_1,Weight_2,'or') plot(Weight_1,Weight_2,'k','linewidth',2) plot(Weight_1',Weight_2','k','linewidth',2) hold off title('t=0'); drawnow;%MATLAB command for flushing the input everytime screen. no=1; do=5; T=10;%number of iteration for self-organising t=1; while (t<=T) n=no*(1-t/T);%for making learning rate decrease with

viii

given given

at output

according

to 10*10 matrix %display(n) d=round(do*(1-t/T)); %display(d) %loop for the 1000 inputs for i=1:1000 Euclidean_distance=(Input_1(i)-Weight_1).^2+(Input_2(i)Weight_2).^2;%minimum distance stored %display(Euclidean_distance) minj1=1; minj2=1; min_norm=Euclidean_distance(minj1,minj2);%initialisig the first 1 x 1 matrix as minimum for reference %display(min_norm) for j1=1:10 for j2=1:10 if Euclidean_distance(j1,j2)<min_norm min_norm=Euclidean_distance(j1,j2);%swapping the values for minimum minj1=j1;%taking the minimum position( may be first column) minj2=j2;%taking the minimum position(may be second is row) end end end min_position_1= minj1;%assigning the minimum element position min_position_2= minj2;%assigning the minimum element position %display(min_position_1 ) %display(min_position_2) %update the winning neuron Weight_1(min_position_1,min_position_2)=Weight_1(min_position_1,min_ position_2)+n*(Input_1(i)- Weight_1(min_position_1,min_position_2)); Weight_2(min_position_1,min_position_2)=Weight_2(min_position_1,min_ position_2)+n*(Input_2(i)- Weight_2(min_position_1,min_position_2)); %update the neighbour neurons for dd=1:1:d% for 3-d plotting %display(dd) min_pos_1=min_position_1-dd;%vertically updating min_pos_2=min_position_2; if (min_pos_1>=1) Weight_1(min_pos_1,min_pos_2)=Weight_1(min_pos_1,min_pos_2)+n*(Input _1(i)-Weight_1(min_pos_1,min_pos_2)); Weight_2(min_pos_1,min_pos_2)=Weight_2(min_pos_1,min_pos_2)+n*(Input _2(i)-Weight_2(min_pos_1,min_pos_2)); end min_pos_1=min_position_1+dd;%vertically updating min_pos_2=min_position_2;

ix

if (min_pos_1<=10) Weight_1(min_pos_1,min_pos_2)=Weight_1(min_pos_1,min_pos_2)+n*(Input _1(i)-Weight_1(min_pos_1,min_pos_2)); Weight_2(min_pos_1,min_pos_2)=Weight_2(min_pos_1,min_pos_2)+n*(Input _2(i)-Weight_2(min_pos_1,min_pos_2)); end min_pos_1=min_position_1; min_pos_2=min_position_2-dd; %horizontally updating if (min_pos_2>=1) Weight_1(min_pos_1,min_pos_2)=Weight_1(min_pos_1,min_pos_2)+n*(Input _1(i)-Weight_1(min_pos_1,min_pos_2)); Weight_2(min_pos_1,min_pos_2)=Weight_2(min_pos_1,min_pos_2)+n*(Input _2(i)-Weight_2(min_pos_1,min_pos_2)); end min_pos_1=min_position_1; min_pos_2=min_position_2+dd;%horizontally updating if (min_pos_2<=10) Weight_1(min_pos_1,min_pos_2)=Weight_1(min_pos_1,min_pos_2)+n*(Input _1(i)-Weight_1(min_pos_1,min_pos_2)); Weight_2(min_pos_1,min_pos_2)=Weight_2(min_pos_1,min_pos_2)+n*(Input _2(i)-Weight_2(min_pos_1,min_pos_2)); end end end

t=t+1; figure(1) plot(Input_1,Input_2,'.b') hold on plot(Weight_1,Weight_2,'or') plot(Weight_1,Weight_2,'k','linewidth',2) plot(Weight_1',Weight_2','k','linewidth',2) hold off title(['t=' num2str(t)]); drawnow; pause(1) end

x

xi

2. Conclusion Self organizing map is one of the most challenging and advanced technology with many application in the field of Science and Research and is being used for many medical purposes. The reason for choosing Kohonen Algorithm is to learn the practical aspect of what is being taught in the classroom as it is a part of the syllabus. One of the main application of this program is to detect tumor in Human brain. We conclude that a Kohonen model is capable of classifying the patients as having glomerular or tubular disease with a high sensitivity and predictive value. The rule-based system does not perform well than the neural networks. The most adequate results were obtained with the hybrid system. Artificial Neural Network models can be used to analyze and predict results much faster for clinical purposes, as it can completely mimic human brain. When Network is trained well it can replace many human functions.

xii

3. References   

www.mathworks.com https://www.youtube.com/watch?v=RqAa8Uu9zbQ

https://en.wikipedia.org/wiki/Self-organizing_map#Applications

xiii