Ideas, CSE 848, Fall 2012

• Benchmarking is quite the rage these days. There's a paper on GP benchmarks (pg. 791) and workshop on black box optimization (companion page). You could review one/all of these papers or try out a new benchmark, indicating why it is a good idea.
• These algorithms are expensive, so there are a number of papers on parallel optimization, especially CUDA. There is a whole Track on Parallel Evolutionary Systems. You could review one of those or implement something along those lines. Is CUDA/GPU worth it? Would MPI be good enough (hey, we have an HPC here)!
• Co-evolution as a top shows up in six different titles. What's the deal with co-evolution? Review/contrast some papers, implement something that compares a "normal" approach and a co-evo approach. Is is worth the effort?
• CMA-ES (Covariance Matrix Adaption-Evolutinary Systems) is a big deal lately. Only one paper, but 6 papers in the Black Box Optimization section and a tutorial! It is a challenging issue (serious math) but explaining it would be useful, and some implementation would be great
• Learning Classifier systems have been around for a long time but have not had the impact that other approaches have. There his a whole Track on on Genetics Based Machine Learning, with 2 best paper awards. Review the state of the art for Learning Classifiers, or talk about CMA-TWEANN (using CMA). Make an application that best uses such a system (what would that be, and why)? Perhaps firewall rules or other network stuff?
• Lots of Application papers in the Application Track. Anything that is along your lines of interest? Good to take a look. Can you do better with a different implementation?
• Hyper-NEAT has been used locally. It shows up a bit in both 2011 and 2012. What is Hyper-NEAT good for and not good for? What would be other implementations you could do for some of the problems they discuss

  ---

  Last modified: Tue Aug 28 17:19:24 EDT 2012