Title Page
1a: Technical Paper (html)
1b: Technical Paper (LaTeX)
1c: Technical Paper (pdf)
2: References
3: Sample Runs
4a: CA Code
4b: GA Code
5: Technical Paper Reading
6: Project Description
7: Oral Report
8a: Daily Logs
8b: Bi-Weekly Goals
8c: Final Iteration Progress Report (#6)
9: Scientific Method
10: Tutorial
11: Next Year
oogleebooglee
|
<![CDATA[
This project was actually done by a student at GMU under Professor Sean Luke,
whom I worked with this summer. It is his attempt to solve the majority
classification problem using cellular automata. Considering we had the same
professor, the projects are fairly similar. To learn more about the project and
hypothesis, see the section on my site that explains the problem.
He is attempting to solve a 149 cell system using 3-neighbor evaluations. In my
project, I used 149 cells, but 7-neighbor evaluations. He also uses a 1-D, circular
(wrap-around), 2-state (binary) system beginning in a randomized state. He runs
the CA for 100 generations and 300 different runs. The final test is on 1000 test
arrays. The breeding options he uses are crossover and mutation.
He sites the previous 5 most successful tests in slight variations on the problem.
The highest accuracy obtained was 82.326%. He also uses Gacs Kurdyomov Lenin's equation, which
he graphs in his paper, as well as citing the Mitchell study in 1993.
He discusses both cellular automatas and genetic algorithms, both of which I
discuss in detail in my techpaper. He sites some old studies and develops his
own variations on the algorithms.
He found that, "the problem in its present form is too restrictive" and recommends
against the use of symmetry.
His project did not appear to be too successful in the end, although he does not
provide solid numbers. I believe that certain constraints that he placed on his
project made it not feasible to be completed. However, it seems as though he identified
most of these mistakes (see above) and so he should be able to improve on the
study.
I learned that, another student attempting the same problem I did, encountered similar,
though different, problems as I. The nature of the problem produces many pitfalls
that can easily be fallen into and hard to get out of. For a majority of the algorithms
and problem approach, we used similar methods. This is encouraging, especially
because these were the sections that worked well.
He has three references in his techpaper.
]]>
|