Recursive percentage based hybrid pattern (RPHP) training for curve fitting
In this paper, we present the RPHP training algorithm, which finds several good local optimal points (pseudo global optima) automatically using an efficient combination of global and local search algorithms. This overcomes the problem of supervised learning algorithms being trapped in a local optima...
محفوظ في:
| المؤلفون الرئيسيون: | , |
|---|---|
| التنسيق: | text |
| اللغة: | English |
| منشور في: |
Institutional Knowledge at Singapore Management University
2004
|
| الموضوعات: | |
| الوصول للمادة أونلاين: | https://ink.library.smu.edu.sg/sis_research/7428 |
| الوسوم: |
إضافة وسم
لا توجد وسوم, كن أول من يضع وسما على هذه التسجيلة!
|
| الملخص: | In this paper, we present the RPHP training algorithm, which finds several good local optimal points (pseudo global optima) automatically using an efficient combination of global and local search algorithms. This overcomes the problem of supervised learning algorithms being trapped in a local optima. Further, to solve a test pattern, we use a modified version of the Kth nearest neighbor (KNN) algorithm as a second level pattern distributor. We tested our approach on three curve fitting problems, whose coefficients were estimated both using genetic algorithms and the RPHP algorithm. The problems were chosen such that they had a small probability of finding a global optimal solution. It was found that the RPHP algorithms performed faster and improved generalization accuracy by as much as 25%. |
|---|