Selective Sampling Using the Query by Committee Algorithm

  • Authors:
  • Yoav Freund;H. Sebastian Seung;Eli Shamir;Naftali Tishby

  • Affiliations:
  • AT&T Labs, Florham Park, NJ 07932. E-mail: yoav@research.att.com;Bell Laboratories, Lucent Technologies, Murray Hill, NJ 07974. E-mail: seung@bell-labs.com;Institute of Computer Science, Hebrew University, Jerusalem, ISRAEL. E-mail: {shamir,tishby}@cs.huji.ac.il;Institute of Computer Science, Hebrew University, Jerusalem, ISRAEL. E-mail: {shamir,tishby}@cs.huji.ac.il

  • Venue:
  • Machine Learning
  • Year:
  • 1997

Quantified Score

Hi-index 0.01

Visualization

Abstract

We analyze the “query by committee” algorithm, a method forfiltering informative queries from a random stream of inputs. Weshow that if the two-member committee algorithm achieves informationgain with positive lower bound, then the prediction error decreasesexponentially with the number of queries. We show that, inparticular, this exponential decrease holds for query learning ofperceptrons.