Lefschetz Center for Dynamical Systems Seminar
Brown Applied Mathematics Pattern Theory and Vision Seminar
Abstract: A decision-making strategy called "active testing" is demonstrated for shape recognition and other tasks in computer vision. The approach is motivated by parlor games such as "Twenty Questions," classical work on sequential testing, and recent work on inductive learning and active vision. I will summarize some completed work (road tracking and handwritten digit recognition) and concentrate on work in progress on focusing and detection. One application is joint work with Chunming Li and Joseph Horowitz on soft tissue classification (and lesion detection) based on magnetic resonance images; the testing strategy is based on a Gaussian data model, which allows for exact entropy calculations and, in principle, accommodates tests at varying resolutions. Another application is joint work with Yali Amit and Bruno Jedynak on detecting faces in cluttered scenes. I will discuss the poor performance of stepwise entropy reduction in the case of rare events, one remedy based on "replicating" globally optimal testing strategies, and analyze the widely-observed efficacy of multiple decision trees in terms of the tradeoff between approximation and estimation error.
Joint Applied Mathematics Colloquium & Brown Statistical Sciences Seminar
Abstract: The presence of variability in the accuracy of radiologists interpreting imaging studies is increasingly recognized as an important aspect of the evaluation of diagnostic technology. This talk will review recent work on approaches to modeling inter-reader variability, including hierarchical regression and Generalized Estimating Equation methods. These approaches are applicable more generally to the analysis of correlated ordinal categorical observations.
PDE/Lefschetz Center Seminar
Department of Mathematics Colloquium
<--- 1997 Index