Description

The problem is specified by the accompanying data file, "vowel.data". This consists of a three dimensional array: voweldata [speaker, vowel, input]. The speakers are indexed by integers 0-89. (Actually, there are fifteen individual speakers, each saying each vowel six times.) The vowels are indexed by integers 0-10. For each utterance, there are ten floating-point input values, with array indices 0-9. The problem is to train the network as well as possible using only on data from "speakers" 0-47, and then to test the network on speakers 48-89, reporting the number of correct classifications in the test set. For a more detailed explanation of the problem, see the excerpt from Tony Robinson's Ph.D. thesis in the COMMENTS section. In Robinson's opinion, connectionist problems fall into two classes, the possible and the impossible. He is interested in the latter, by which he means problems that have no exact solution. Thus the problem here is not to see how fast a network can be trained (although this is important), but to maximise a less than perfect performance.

Related Papers

  • M. Layton and M. J. F Gales. CAMBRIDGE UNIVERSITY ENGINEERING DEPARTMENT Maximum Margin Training of Generative Kernels. CUED. 2004. [link]
  • Matthew Brand. Pattern discovery via entropy minimization. MERL -- A MITSUBISHI ELECTRIC RESEARCH LABORATORY. 1998. [link]
  • [link]
  • [link]
  • [link]
  • [link]

Related datasets