Learning changing concepts by exploiting the structure of change

被引:30
作者
Bartlett, PL [1 ]
Ben-David, S
Kulkarni, SR
机构
[1] Australian Natl Univ, Res Sch Informat Sci & Engn, Canberra, ACT 0200, Australia
[2] Technion Israel Inst Technol, Dept Comp Sci, IL-32000 Haifa, Israel
[3] Princeton Univ, Dept Elect Engn, Princeton, NJ 08544 USA
基金
澳大利亚研究理事会;
关键词
computational learning theory; sample complexity; concept drift; estimation; tracking;
D O I
10.1023/A:1007604202679
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
This paper examines learning problems in which the target function is allowed to change. The learner sees a sequence of random examples, labelled according to a sequence of functions, and must provide an accurate estimate of the target function sequence. We consider a variety of restrictions on how the target function is allowed to change, including infrequent but arbitrary changes, sequences that correspond to slow walks on a graph whose nodes are functions, and changes that are small on average, as measured by the probability of disagreements between consecutive functions. We first study estimation, in which the learner sees a batch of examples and is then required to give an accurate estimate of the function sequence. Our results provide bounds on the sample complexity and allowable drift rate for these problems. We also study prediction, in which the learner must produce online a hypothesis after each labelled example and the average misclassification probability over this hypothesis sequence should be small. Using a deterministic analysis in a general metric space setting, we provide a technique for constructing a successful prediction algorithm, given a successful estimation algorithm. This leads to sample complexity and drift rate bounds for the prediction of changing concepts.
引用
收藏
页码:153 / 174
页数:22
相关论文
共 15 条
[1]  
Anthony M., 1999, Neural network learning: theoretical foundations, Vfirst
[2]  
Bartlett P. L., 1992, Proceedings of the Fifth Annual ACM Workshop on Computational Learning Theory, P243, DOI 10.1145/130385.130412
[3]  
BARTLETT PL, 1996, LEARNING CHANGING PR
[4]   On the complexity of learning from drifting distributions [J].
Barve, RD ;
Long, PM .
INFORMATION AND COMPUTATION, 1997, 138 (02) :170-193
[5]  
Bertoni A., 1992, Proceedings of the Fifth Annual ACM Workshop on Computational Learning Theory, P265, DOI 10.1145/130385.130414
[6]  
Blum A., 1992, Proceedings of the Fifth Annual ACM Workshop on Computational Learning Theory, P231, DOI 10.1145/130385.130411
[7]   LEARNABILITY AND THE VAPNIK-CHERVONENKIS DIMENSION [J].
BLUMER, A ;
EHRENFEUCHT, A ;
HAUSSLER, D ;
WARMUTH, MK .
JOURNAL OF THE ACM, 1989, 36 (04) :929-965
[8]  
CONNOLLY AJ, 1992, CONTR 92 C P I ENG A
[9]  
Freund Y., 1997, Computational Learning Theory. Third European Conference, EuroCOLT '97. Proceedings, P109
[10]   A GUIDED TOUR OF CHERNOFF BOUNDS [J].
HAGERUP, T ;
RUB, C .
INFORMATION PROCESSING LETTERS, 1990, 33 (06) :305-308