Stability Analysis of ECOC Kernel Machines
| Autor: | Xiaolong Fu, Aijun Xue, Xiaodan Wang |
|---|---|
| Rok vydání: | 2017 |
| Předmět: |
Pointwise
Generalization Property (programming) Computer science Model selection Open problem Stability (learning theory) 02 engineering and technology 01 natural sciences Kernel method Kernel (statistics) 0103 physical sciences 0202 electrical engineering electronic engineering information engineering 020201 artificial intelligence & image processing 010306 general physics Algorithm |
| Zdroj: | Lecture Notes in Computer Science ISBN: 9783319715889 ICIG (2) |
| Popis: | Error correcting output codes kernel machines (ECOC kernel machines) are ensemble of kernel machines based on ECOC decomposition methods. How to improve the generalization capability of this framework is an open problem. In this paper, we discussed the condition for generalization in terms of the stability property of ECOC kernel machines. Here we provide a proof for the result that an ECOC kernel machine has the pointwise hypothesis stability. This stability property can be calculated by training on the training dataset once and has clear and meaningful formulation. It can be applied to tune the kernel parameters in model selection and design good matrixes for ECOC kernel machines. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|