Most accurate predictions are typically obtained by learning machines with complex feature spaces (as e.g. induced by kernels). Unfortunately, such decision rules are hardly access...
Most existing web video search engines index videos by file names, URLs, and surrounding texts. These types of video roughly describe the whole video in an abstract level without ...
Selecting the optimal kernel is an important and difficult challenge in applying kernel methods to pattern recognition. To address this challenge, multiple kernel learning (MKL) ...
Adaptation techniques based on importance weighting were shown effective for RankSVM and RankNet, viz., each training instance is assigned a target weight denoting its importance ...
This paper describes the development of a new document ranking system based on layout similarity. The user has a need represented by a set of ”wanted” documents, and the syste...
May Huang, Daniel DeMenthon, David S. Doermann, Ly...