Computer Science Department
School of Computer Science, Carnegie Mellon University
Using Discard-based Search for Indexed Search
Mahadev Satyanarayanan, Christine Henderson*,
Automated indexing of complex data such as images remains a challenging problem today in spite of extensive research. An alternative approach, called discard-based search, uses code fragments called searchlets to perform content-based computation in response to a specific query. In this paper, we describe a new, two-phased usage model for discard-based search. In the first-phase, human experts use discard-based search to create searchlets that reflect their classification expertise. In the second phase, these searchlets are used to preprocess complex data for indexing. This hybrid approach preserves the positive characteristics of indexed search, while offering flexibility in the creation of searchlets and in tuning their precision-recall characteristics. Most importantly, it offers ample opportunity for human expertise and new knowledge to be efficiently incorporated into the process of indexing complex data.
*University of Pittsburgh Medical Center (UPMC)