Assignment # 3
(Posted on March 9 - Due: 5 PM on May 7, 2012
Marks: 10

Train a Neural Network using the data set provided in the following file.

The data set represent the KSE Index values during 2004 to 2006. There are four columns in the file, namely, Index(t-3), Index(t-2), Index(t-1) and Index(t), represented as X1, X2, X3 and Y, respectively. Your job is to learn Index(t) as a function of (Index(t-1), Index(t-2) and Index(t-3)). In other words, your job is to predict the value of tomorrow's index value using the values of today, yesterday and day before yesterday.

The training of the neural network should be done in the following manner:

(i) Backpropagation
(ii) Weights Evolution using either EA or PSO (Evolutionary Neural Network)

Plot the values of the original index(t), predicted index(t) using Backpropagation and Evolutionary Neural Network. Also report the correlation between the actual and predicted values (of both approaches) and the time taken by both approaches. Describe your findings.

Assignment # 2
(Posted on March 17, 2013 - Due: 2 AM on April 16, 2013 )
Marks: 4

You need to present an application of Computational Intelligence. I would recommend that you select a paper on EC and summarize its finding in the class. The application should use one of the techniques we have discussed in the class (EA, PSO, ACO, NN, etc.).

Assignment # 1 (c)
(Posted on March 17, 2013 - Due: 2 AM on April 2, 2013 )
Marks: 4

Compare two combinations of EA (a: FPS + Truncation, b: RBS + BT) against Particle Swarm Optimization (PSO) and Artificial Immune Systems. Keep the population size (mu) as 10 for all the algorithms.

The functions used in this part are:

a) Rosenbrock function
b) Himmelblau's function

In addition to drawing average-best-so-far curves and average-average fitness curves, also draw average population variance curve. Furthermore, do hypothesis testing to compare results of PSO an AIS.

Assignment # 1 (b)
(Posted on March 5, 2013 - Due: 2 AM on March 24, 2013 )
Marks: 4

Compare two combinations of EA (a: FPS + Truncation, b: RBS + BT) against canonical versions of Evolutionary Programming and Evolution Strategy. For Evolution Strategy, use (mu, lambda) version. Keep the population size (mu) as 10 for all the algorithms. Lambda should be set to 15 for ES. Details of the algorithms are as follows:

Parent Selection: Deterministic (each parent produces an offspring)
Crossover: None
Mutation: Gaussian (for every gene - no 20% rule)
Survival Selection: Probabilistic (either FPS or RBS) on 20 solutions (10 parents + 10 offspring)

Parent Selection: Uniform Stochastic (or binary tournament)
Crossover: Discrete/Average
Mutation: Gaussian (20% rule applicable)
Survival Selection: Truncation only applied on 16 offspring

The functions used in this part are:

a) Function # 1 of Part (a)
b) Rosenbrock function (Function # 2) of Part (a)
b) Himmelblau's function
where ranges of x and y are: -4 < x, y < 4

In addition to providing comparisons based on average-best-so-far and average-average-fitness curves, you also need to do hypothesis testing between the best two combinations.

Assignment # 1 (a)
(Posted on Feb 21, 2013 - Due: 2 AM on March 10, 2013 )
Marks: 4

Using the functions provided in Unit # 4, compare the performance of different combination of selection schemes in an EA. The following schemes should be considered:
Parent Selection: Fitness Proportional, Rank-based and Binary Tournament
Survival Selection: Truncation and Binary Tournament

Each combination should be run at least 10 times and the evaluation should be performed using:
(i) average best-so-far curve and
(ii) average average-population fitness curve

You will present your findings in the class. You are welcome to try your own ideas as well in addition to the tasks described above.