On the complexity of learning from drifting distributions

被引:11
作者
Barve, RD [1 ]
Long, PM [1 ]
机构
[1] NATL UNIV SINGAPORE,ISCS DEPT,SINGAPORE 119260,SINGAPORE
关键词
D O I
10.1006/inco.1997.2656
中图分类号
TP301 [理论、方法];
学科分类号
081202 ;
摘要
We consider two models of on-line learning of binary-valued functions from drifting distributions due to Bartlett. We show that if each example is drawn from a joint distribution which changes in total variation distance by at most O(epsilon(3)/(d log(1/epsilon))) between trials, then an algorithm can achieve a probability of a mistake at most epsilon worse than the best function in a class of VC-dimension d. We prove a corresponding necessary condition of O(epsilon(3)/d). Finally, in the case that a fixed function is to be learned from noise-free examples, we show that if the distributions on the domain generating the examples change by at most O(epsilon(2)/(d log(1/epsilon))), then any consistent algorithm learns to within accuracy epsilon. (C) 1997 Academic Press.
引用
收藏
页码:170 / 193
页数:24
相关论文
共 34 条
[1]  
Angluin D., 1988, Machine Learning, V2, P343, DOI 10.1007/BF00116829
[2]  
[Anonymous], P 36 ANN S FDN COMP
[3]  
ANTHONY M, 1992, CAMBRIDGE TRACTS THE, V30
[4]  
ANTHONY M, 1990, 1990 WORKSH COMP LEA, P246
[5]  
ARORA S, 1993, AN S FDN CO, P724
[6]  
BARTLETT PL, 1996, 1995 C COMP LEARN TH, P131
[7]  
BARTLETT PL, 1995, UNPUB PREDICTION LEA
[8]  
BARTLETT PL, 1992, 1992 WORKSH COMP LEA, P243
[9]  
BARTLETT PL, 1995, UNPUB
[10]  
Ben-David S., 1989, Proceedings of the Second Annual Workshop on Computational Learning Theory, P285