We show that with a uniform prior on models having the same training error, early stopping at some fixed training error above the training error minimum results in an increase in the expected generalization error.

This content is only available as a PDF.
© 1999 Massachusetts Institute of Technology
1999
You do not currently have access to this content.