anytime algorithms for learning anytime classifiers saher ... - Technion

Technion - Computer Science Department - Ph.D. Thesis PHD-2008-12 - 2008
Gogan, 2004). If we chose to update the portfolio at specific time points,
we would like the learner to exploit the time between these updates.
Furthermore, researchers in the field can benefit from the automatic method
for cost assignments we have developed. Only a few UCI datasets have assigned
costs. In this work we designed a semi-randomized method for assigning costs
to existing datasets. We applied this method on 25 datasets and established a
repository, available at: http://www.cs.technion.ac.il/∼esaher/cost.
This research can be extended in several directions. We intend to apply
monitoring techniques for optimal scheduling of the anytime learners. We also
plan to use different measures for tree quality and compare their utility. While
the tree size and expected error were generally successful, our sampling approach
did not, in few cases, yield significant improvement. Using other measures may
improve the performance in these cases.
We also intend to test the performance of our framework on other cost schemes
that involve other types of cost. We believe that the generality of our framework
will allow excellent results to be obtained under other setups as well. To reduce
the runtime of our anytime algorithms, we plan to cache some of the lookahead
trees and use them, rather than resampling at each node. If a split is chosen,
the sample of already available subtrees can be used to evaluate its descendants
as well. Finally, an important advantage of our method is that it can be easily
parallelized. Assume, for example, that we decided on samples of size r. Then, r
different machines can independently form the sample and speed up the induction
process by a factor of r. We intend to consider this direction in the future.
138