An initial study of an adaptive hierarchical vision system

Marcus A. Maloof

We describe an empirical study of an adaptive, hierarchical vision system. Using a simple vision task requiring both low-level and high-level processing, we examined how three schemes of feedback for on-line learning affected the true positive rate, the number of instances used for learning, and the need for user feedback. The first scheme used for learning those instances for which the user provided feedback. The second used all instances, assuming that no feedback meant correct classification. In the end, a hybrid scheme with each method at different levels yielded the best results, showing that more examples for learning significantly improved the true positive rate of the classifiers at the lower level, but not at the higher level. Furthermore, this hybrid method did not increase the need for user feedback.

Paper available in PostScript (gzipped) and PDF.

Slides from the talk available in PostScript (gzipped) and PDF.

Maloof, M.A. (2000). An initial study of an adaptive hierarchical vision system. Proceedings of the Seventeenth International Conference on Machine Learning, 567–573. San Francisco, CA: Morgan Kaufmann.

@inproceedings{maloof.icml.00,
  author = "Maloof, M.A.",
  title = "An initial study of an adaptive hierarchical vision system",
  booktitle = "{Proceedings of the Seventeenth International Conference on
     Machine Learning}",
  year = 2000,
  pages = "567--573",
  publisher = "Morgan Kaufmann",
  address = "San Francisco, CA"
}