A predictor-corrector affine scaling method to train optimized extreme learning machine |
| |
| Institution: | 1. College of Information Engineering, Nanjing University of Finance and Economics, Nanjing 210023, China;2. CAS Key Laboratory of Planetary Sciences, Purple Mountain Observatory, Chinese Academy of Sciences, Nanjing 210023, China;3. NARI Group Corporation (State Grid Electric Power Research Institute), Nanjing 211106, China;1. School of Electrical Engineering and Automation, Jiangsu Normal University, Xuzhou 221116, China;2. Department of Electrical Engineering, Yeungnam University, 280 Daehak-Ro, Gyongsan 38541, Republic of Korea;3. School of Automation, Nanjing University of Science and Technology, Nanjing 210094, China;1. Department of Electronics and Telecomunication Engineering, State University of Rio de Janeiro (DETEL/UERJ), Rio de Janeiro, Brazil;2. Department of Electrical Engineering, Federal University of Rio de Janeiro (COPPE/UFRJ), Rio de Janeiro, Brazil;1. Department of Electrical Engineering, University of Technology of Paraná – UTFPR Cornélio Procópio 86300-000, PR, Brazil;2. Federal Institute of São Paulo – IFSP, Campus Avaré, Avaré 18707-150, SP, Brazil;1. Department of Aerospace Engineering, University of Bristol, Bristol BS8 1TR, United Kingdom;2. European Space Agency, ESTEC, Noordwijk 2201AZ, the Netherlands |
| |
| Abstract: | Optimized extreme learning machine (OELM) has been shown to achieve high performance on classification problems due to its simple dual form. This paper presents a predictor-corrector affine scaling interior point method to exploit the dual problem of OELM. This method aims to combine a predictor step with a corrector step for determining the descent Newton direction. At each iteration, the predictor step focuses on the complementarity gap reduction and computes an affine scaling direction to estimate the extent of the reduction of complementarity gap, while the corrector step traces the central path towards the optimal solution by high order approximation, and computes the corresponding center direction. Then, the Newton direction is combined by using both two directions, and the iteration sequence of interior feasible points converges to the optimal solution. Extensive experimental evaluations on various benchmark datasets show that the proposed algorithms outperform other interior point-based or active set-based algorithms. Moreover, they are able to converge in fewer iterations, which are independent of kernel type, dataset size and dimensionality. |
| |
| Keywords: | |
| 本文献已被 ScienceDirect 等数据库收录！ |
|
