Genetic Programming Assignment
Evaluation: An 8-12-page paper, and relevant code and data listings.
Topics: Two types of topics are as follows:
1. Application: Use GP to solve some application problem of your choosing. You might try using a new variation of GP towards a known problem, or even an improvement of some other researcher's approach as described in the literature. You might even think of a novel problem in which GP has yet to be applied. Lots of data for different real-world problems is available at the UCI Machine Learning Repository (www.ics.uci.edu/~mlearn/MLRepository.html).
The results of your experiment should be described in a suitable report, which lists all the relevant parameters and results. Attention should be given to experimental methodology and analyses. Your assignments will give you practice in writing formal reports.
2. System: Create a new GP system with some interesting features, for example:
- Koza's tree-based GP system in a language of your choice (other than Lisp). You might embellish it with some new features.
- Extend an existing GP system (ECJ, lil-GP 1.1, OpenBeagle, ...) in a useful and nontrivial way.
- Create a visualization tool for a GP system. Many papers exist on techniques for visualizing runs, populations, progress, etc.
- Applying GP on a GPU (e.g. using CUDA or OpenCL).
- A new GP representation, using grammars, linear chromosomes, ...
System projects will require testing on experimental data. Therefore, such projects will still require a formal report of experimental evidence of the implemented system. There will be less emphasis on experimental methodology and analysis compared to application-oriented projects. The report should also describe the system design and implementation of the implemented system.