On the value of partial information for learning from examples

被引:9
|
作者
Ratsaby, J [1 ]
Maiorov, V
机构
[1] Technion Israel Inst Technol, Dept Elect Engn, IL-32000 Haifa, Israel
[2] Technion Israel Inst Technol, Dept Math, IL-32000 Haifa, Israel
关键词
D O I
10.1006/jcom.1997.0459
中图分类号
TP301 [理论、方法];
学科分类号
081202 ;
摘要
The PAC model of learning and its extension to real valued function classes provides a well-accepted theoretical framework for representing the problem of learning a target function g(x) using a random sample {(x(i), g(x(i)))}(i=1)(m). Based on the uniform strong law of large numbers the PAC model establishes the sample complexity, i.e., the sample size m which is sufficient for accurately estimating the target function to within high confidence. Often, in addition to a random sample, some form of prior knowledge is available about the target. It is intuitive that increasing the amount of information should have the same effect on the error as increasing the sample size. But quantitatively how does the rate of error with respect to increasing information compare to the rate of error with increasing sample size? To answer this we consider a new approach based on a combination of information-based complexity of Traub et al. and Vapnik-Chervonenkis (VC) theory. In contrast to VC-theory where function classes of finite pseudo-dimension are used only for statistical-based estimation, we let such classes play a dual role of functional estimation as well as approximation. This is captured in a newly introduced quantity, rho(d)(F), which represents a nonlinear width of a function class F. We then extend the notion of the nth minimal radius of information and define a quantity I-n,I-d(F) which measures the minimal approximation error of the worst-case target g is an element of F by the family of function classes having pseudo-dimension d given partial information on g consisting of values taken by n linear operators. The error rates are calculated which leads to a quantitative notion of the value of partial information for the paradigm of learning from examples. (C) 1997 Academic Press.
引用
收藏
页码:509 / 544
页数:36
相关论文
共 50 条
  • [21] LEARNING FROM EXAMPLES.
    Pawlak, Zdzislaw
    Bulletin of the Polish Academy of Sciences: Technical Sciences, 1986, 34 (9-10): : 573 - 586
  • [22] Learning from Cluster Examples
    Toshihiro Kamishima
    Fumio Motoyoshi
    Machine Learning, 2003, 53 : 199 - 233
  • [23] Attention, Learning, and the Value of Information
    Gottlieb, Jacqueline
    NEURON, 2012, 76 (02) : 281 - 295
  • [24] Learning from order examples
    Kamishima, T
    Akaho, S
    2002 IEEE INTERNATIONAL CONFERENCE ON DATA MINING, PROCEEDINGS, 2002, : 645 - 648
  • [25] CONCEPT-LEARNING FROM EXAMPLES AND COUNTER EXAMPLES
    RALESCU, AL
    BALDWIN, JF
    INTERNATIONAL JOURNAL OF MAN-MACHINE STUDIES, 1989, 30 (03): : 329 - 354
  • [26] SOME EXAMPLES OF NONCLASSICAL BOUNDARY VALUE PROBLEMS FOR PARTIAL DIFFERENTIAL EQUATIONS
    BEREZANSKII, IM
    DOKLADY AKADEMII NAUK SSSR, 1960, 131 (03): : 478 - 481
  • [27] Exploiting counter-examples for active learning with partial labels
    Zhang, Fei
    Ye, Yunjie
    Feng, Lei
    Rao, Zhongwen
    Zhu, Jieming
    Kalander, Marcus
    Gong, Chen
    Hao, Jianye
    Han, Bo
    MACHINE LEARNING, 2024, 113 (06) : 3849 - 3868
  • [28] Exploiting counter-examples for active learning with partial labels
    Fei Zhang
    Yunjie Ye
    Lei Feng
    Zhongwen Rao
    Jieming Zhu
    Marcus Kalander
    Chen Gong
    Jianye Hao
    Bo Han
    Machine Learning, 2024, 113 : 3849 - 3868
  • [29] Communication-learning-information technology applied examples
    Appelberg, L
    INFORMATION TECHNOLOGY: SUPPORTING CHANGE THROUGH TEACHER EDUCATION, 1997, : 216 - 221
  • [30] INFORMATION AND COMMUNICATION TECHNOLOGY AS A LINK BETWEEN INFORMAL AND FORMAL LEARNING - EXAMPLES FROM CROATIA
    Rogic, Ana Marija
    Dimic, Jasmina Vrkic
    EDULEARN14: 6TH INTERNATIONAL CONFERENCE ON EDUCATION AND NEW LEARNING TECHNOLOGIES, 2014, : 5387 - 5396