Assignment 2 (Neural Computation, Spring 2009)

Number of problems/points: Five problems for total of 100 points

Out: February 18, 2009

Due: February 25, in class.

Problem 1: (25 points)

Implement the Pocket algorithm and use it to solve a double-moon classification problem from Problem 2 of Homework assignment 1 for d= -1.

Problem 2: (25 points)

Repeat tasks defined at Problem 1 using a two layer neural network with 4 hidden units trained using the backpropagation algorithm.

Obtain the Wisconsin Prognostic Breast Cancer dataset from the UCI Repository of Machine Learning Databases at http://www.ics.uci.edu/~mlearn/MLRepository.html

and perform 5 cross-validation experiments to estimate the precision accuracy

of a feedforward neural network trained using the backpropagation algorithm

on these problems:

Problem 3: (25 points) 
Prediction of recurrence of cancer at 24 months (field 2 in dataset).
Problem 4: (25 points)
Prediction of time to recur (field 3 in recurrent records)

In problems 3 and 4 consider optimizing the number of hidden neurons, learning rates and momentum parameters. In addition to accuracy, report the average number of iterations needed for convergence and the standard deviation.