Probabilistic Ontology Induction for Generalized Video Understanding
Final rept. 5 Mar 2009-30 Jun 2010
STATE UNIV OF NEW YORK COLL AT BUFFALO
Pagination or Media Count:
The initial plan in his Phase I research was to induce a probabilistic ontology directly from video of a particular phenomenon. The human role was left to the end to supply the semantics. However, he has realized that the human needs to be involved throughout the process. This has led him to investigate a new branch of clustering called active clustering, which involves the human throughout the clustering process. Ultimately patterns are discovered in the data, which can then be used for model building and inference, but few or no assumptions are made beforehand and the human expert is involved throughout the entire process. This is a new, unexplored, and unpublished set of ideas that is the topic of his Phase II project, which was funded and began in late June 2010.