|
| 摘要: |
| 本文研究了多阈值和非一致分布下的在线分位数回归算法，在每一次迭代中，样本会来自不同的分布和取不同的阈值.利用边缘分布在对偶空间中多项式收敛的性质，我们得到了算法的学习速度，并且做了相应的数值模拟来支持我们的结论. |
| 关键词: 非一致分布 在线学习 分位数回归 再生核希尔伯特空间 |
| DOI： |
| 分类号:O29 |
| 基金项目:Supported by National Natural Science Foundation of China(11671307). |
|
| ONLINE QUANTILE REGRESSION WITH VARYING THRESHOLDS AND NON-IDENTICAL SAMPLING DISTRIBUTIONS |
|
JIANG Ming-qin
|
| Abstract: |
| In this paper we study the online quantile regression algorithm with varying thresholds and non-identical sampling distributions, where at each time a sample is drawn independently from different probability distributions and the threshold values decrease with the iteration process. The learning rate of the algorithm is obtained under the assumption that the sequence of marginal distribution converges polynomially fast in the dual of a Hölder space. Several numerical simulations are presented to support our results. |
| Key words: sampling with non-identical distributions online learning quantile regression ϵ-insensitive pinball loss reproducing kernel Hilbert spaces |
