Paul Fischer, Norbert Klasner:
For applications in Machine Learning and forecasting the Minimum
Disagreement problem plays a fundamental role. This is the
combinatorial problem of selecting a best explanation for observed
data from a set of possible ones. It has recently been shown that the
Minimum Disagreement strategy can be outperformed by randomized
strategies if the observed data is noisy. This new strategy amounts
to finding not a single best explanation for the data, but to finding
a number of good explanations which satisfy a pairwise independence
condition. Here we show a general approach for solving this new
combinatorial problem. We also show that for some classes the problem
can be solved more efficiently by a problem specific approach.
Last update: 11/17/98. Norbert Klasner, klasner@lmi.ruhr-uni-bochum.de.
Return to list of my publications.