Interactive l earning using a \society o f m odels"

R Picard
unpublished
Digital library access is driven by features, but features are often context-dependent a nd noisy, and their relevance for a query is not always obvious. This paper describes an approach f o r utilizing many d ata-dependent, user-dependent, and task-dependent features in a semi-automated tool. Instead of requiring universal similarity measures or manual selection of relevant f e a t ures, the approach p r o vides a learning algorithm for selecting and combining groupings of the data, where
more » ... ings can be induced by h i g h l y s p ecialized and context-dependent f eatures. The selection process is guided by a r i c h example-based interaction with the user. The inherent combinatorics of using multiple features is reduced by a m ultistage grouping generation, weighting, and collection process. The stages closest to the user are trained fastest and slowly propagate their adaptations back t o earlier stages. The weighting stage adapts the collection stage's search s pace across uses, so that, in later interactions, good groupings are found given few examples from the user. Described is an interactive-time implementation of this architecture for semi-automatic within-image segmentation and across-image labeling, driven by c o ncurrently active c olor models, texture models, or manually-provided groupings.
fatcat:bsdd77qzfvhytas5myn6azcc74