The Canonical Distortion Measure in Feature Space and 1-NN Classification

@inproceedings{Baxter1997TheCD,
  title={The Canonical Distortion Measure in Feature Space and 1-NN Classification},
  author={Jonathan Baxter and Peter L. Bartlett},
  booktitle={NIPS},
  year={1997}
}
We prove that the Canonical Distortion Measure (CDM) [2, 3] i s the optimal distance measure to use for 1 nearest-neighbour (1NN) classification, and show that it reduces to squared Euclidean distan ce in feature space for function classes that can be expressed as linear co mbinations of a fixed set of features. PAC-like bounds are given on the sam plecomplexity required to learn the CDM. An experiment is prese nted in which a neural network CDM was learnt for a Japanese OCR envir o ment and… CONTINUE READING

References

Publications referenced by this paper.
Showing 1-5 of 5 references

Cherry Blossom: A Syste m for Reading Unconstrained Handwritten Page Images

  • S. N. Srihari, T. Hong, Z. Shi
  • In Symposium on Document Image Understanding…
  • 1997
1 Excerpt

Similar Papers

Loading similar papers…