Distributed Function Mining for Gene Expression Programming Based on Fast Reduction |
| |
Authors: | Song Deng Dong Yue Le-chan Yang Xiong Fu Ya-zhou Feng |
| |
Affiliation: | 1.Institute of Advanced Technology, Nanjing University Post & Telecommunication, Nanjing, 210023, China;2.International Institute for Earth System Science, Nanjing University, Nanjing, 210093, China;3.School of Computer, Nanjing University Post & Telecommunication, Nanjing, 210023, China;Cedars-Sinai Medical Center, UNITED STATES |
| |
Abstract: | For high-dimensional and massive data sets, traditional centralized gene expression programming (GEP) or improved algorithms lead to increased run-time and decreased prediction accuracy. To solve this problem, this paper proposes a new improved algorithm called distributed function mining for gene expression programming based on fast reduction (DFMGEP-FR). In DFMGEP-FR, fast attribution reduction in binary search algorithms (FAR-BSA) is proposed to quickly find the optimal attribution set, and the function consistency replacement algorithm is given to solve integration of the local function model. Thorough comparative experiments for DFMGEP-FR, centralized GEP and the parallel gene expression programming algorithm based on simulated annealing (parallel GEPSA) are included in this paper. For the waveform, mushroom, connect-4 and musk datasets, the comparative results show that the average time-consumption of DFMGEP-FR drops by 89.09%%, 88.85%, 85.79% and 93.06%, respectively, in contrast to centralized GEP and by 12.5%, 8.42%, 9.62% and 13.75%, respectively, compared with parallel GEPSA. Six well-studied UCI test data sets demonstrate the efficiency and capability of our proposed DFMGEP-FR algorithm for distributed function mining. |
| |
Keywords: | |
|
|