The multistage combinatorial GMDH algorithm for parallel processing of high-dimensional data. Oleksiy Koshulko, Anatoliy Koshulko

Abstract. Combinatorial algorithm of the Group method of data handling (GMDH) is a powerful tool for building noise resistant multidimensional mathematical models with optimal complexity. Full combinatorial search is a preferable type of model search, but it can not be applied with high-dimensional objects because of high computational complexity. However there is a way to decrease the computational complexity by suggesting a reasonable procedure of partial combinatorial search. For this purpose we compare simple complexity limitations of GMDH models with a procedure called Multistage combinatorial algorithm.

Keywords. Combinatorial GMDH, parallel processing, high-dimensional data.

References.

1. Madala H.R., Ivakhnenko A.G.: Inductive Learning Algorithms for Complex Systems Modeling. CRC Press, 368p., 1994.

2. Ivakhnenko A.G., Stepashko V.S.: Noise-immunity of modeling. Kiev: Naukova dumka, 216 p., 1985. (in Russian)

3. O.Koshulko A.Koshulko. Adaptive parallel implementation of the combinatorial GMDH algorithm. In proceedings of International Workshop on Inductive Modelling, Prague, 2007.

4. Top500 List –  http://www.top500.org

5. HPL Benchmark –  http://www.netlib.org/benchmark/hpl/