Error-corrective Feature Extraction in Handwritten Digit Recognition

@inproceedings{Laaksonen1997ErrorcorrectiveFE,
  title={Error-corrective Feature Extraction in Handwritten Digit Recognition},
  author={Jorma Laaksonen and Erkki Oja},
  year={1997}
}
The idea of feedback and error-correction is central in neurally motivated classiication algorithms. Most of the neural models, however, take the preceding feature extraction stage as given. Unfortunately, essential information may be unrecoverably lost in the feature extraction phase, leading to degraded classiication accuracy. Enhanced overall classiication performance would follow, if the neural adaptivity could be extended from the classiier to the feature extractor. In our work, we have… CONTINUE READING