A decision tree's classification performance can drop if the tree is used in a changed context such as different accent in speach recognition. This brittleness can partially be rectified by the use of a cheap second tier implemented as a linear classifier. The transfer of the tree to a novel context is accomplished by re-inducing the second tier, without the need to re-induce the more expensive first tier. Experiments reported in this paper indicate that quick adaptation to the target context can indeed be achieved.
|Original language||English (US)|
|Number of pages||10|
|State||Published - Jun 1 2000|
ASJC Scopus subject areas
- Theoretical Computer Science
- Computer Science Applications
- Artificial Intelligence