Scribd
Upload
138 views

KOM6110 ANN - Machine - Learning Assignment 1 Spring 2017

The document outlines an assignment for an artificial neural network machine learning course involving training perceptrons and multilayer perceptrons on synthetic datasets to classify samples. For part 1, students are asked to train a perceptron using gradient descent and stochastic gradient descent algorithms to classify samples from 2 classes. For part 2, students must create and train a multilayer perceptron using backpropagation to classify samples from 3 classes. The document provides details on the datasets, required tasks, and submission instructions.

Uploaded by

Kamer Bibuçukvolt Semercioglu
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
138 views

KOM6110 ANN - Machine - Learning Assignment 1 Spring 2017

The document outlines an assignment for an artificial neural network machine learning course involving training perceptrons and multilayer perceptrons on synthetic datasets to classify samples. For part 1, students are asked to train a perceptron using gradient descent and stochastic gradient descent algorithms to classify samples from 2 classes. For part 2, students must create and train a multilayer perceptron using backpropagation to classify samples from 3 classes. The document provides details on the datasets, required tasks, and submission instructions.

Uploaded by

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

1 KOM6110 ANN Machine Learning

Assignment 1 Due Date: April.12.2017

1- Given two synthetic datasets


class2_tr : 220 samples with 2 features and 1 label bit, from 2 classes
class2_test: 80 samples with 2 features and 1 label bit, from 2 classes

a) Create a suitable perceptron (a single unit) structure


b) Train it with delta rule which uses gradient descent, and test it with the test data.
c) Train it with delta rule which uses stochastic gradient descent, and test it with the test data
d) Get related figures to display data distribution, classification for training data, classification for test
data, training error vs iterations

You have to write .m files to train the perceptron with


delta rule using gradient descent
delta rule using stochastic gradient descent

2- Given two synthetic datasets


class3_tr : 330 samples with 2 features and 2 label bits, from 3 classes
class3_test: 120 samples with 2 features and 2 label bits, from 3 classes
(The labels are given yet we assume that we do not know them)

a) Create a suitable MLP structure


b) Train it with backpropagation algorithm using the training data, and test it using the test data
c) Get related figures to display data distribution, classification for training data, classification for test
data, training error vs iterations

You can consider the classification result of sample good if the output is close enough to the target
value. If the desired value is = [1 1] and you have = [0.99 0.93] you may regard it still a good
classification
The weights are initialized randomly for each run you may have different results
You can do different experiments on a) number of hidden layer units, b) learning rate, c) momentum
You have to write .m files to train and test the MLP you created.

3- Create a doc file as the documentation about what you may obtain, and write.
Dr. Muharrem Mercimek

a. The due date is firm and it is the midnight of April.12.2017


b. Submit your document as a {.doc}or a {.pdf} document in which the figures and codes are included. Also,
submit your .m files. Name your document as KOM6110_yourName_yourNumber_Assignment1.{doc,pdf}
c. Zip your files and name the zip file KOM6110_yourName_yourNumber_Assignment1.{zip,rar}
d. Overall, you have to put 2 kind of files .doc or .pdf and .m files in one zip file.
e. Give an appropriate title name (e.g: KOM6110_yourName_yourNumber_Assignment1) for the e-mail.
f. Send your email to [email protected]

You might also like