Thursday, May 03, 2007

New article

Our work on stroke type diagnostics is almost completed. Dr. Rebrova should arrange with stroke-center about experiments and only after them we can continue work. Rather they will continue this work without me. I just have no time for this. Maybe later. inshaAllah.
Article describes the problems which present in diagnostics of type of stroke, and ours approach which is right-diagnose more than in 97%. Rebrovas part was to collect and reduce input data dimensionality via statistical methods. Redko and Mosalov, chose the instance of multi-layer neural network via training a huge number of agents and applying genetic algorithms to them. After MNN was chosen, they get some result on it in 3-deminseional space (each dimension corresponds to type of stroke). But result weren’t good enouhg yet. And I decided to train Kohonen NN on that 3-dimensianl space. Results are really great. So, if anyone is interesting, I can send full PDF of this article.

No comments: