演講者：張洛賓教授

單　位：國立交通大學應用數學系

日　期：2012年5月2日　PM 14:30

地　點：國立高雄大學理學院408室

講　題：Context, Computation, and Asymptotic ROC Performance in Hierarchical Models

摘　要：

Under any reasonable formulation, the vision problem is provably NP-hard. So are many other problems from biological systems, e.g. protein folding and neuron signal transmission. The nature handles these problems routinely, but how does the nature compute? One of the common problems is the combinatorial complexity. In this talk, I will propose a probabilistic combinatorial formulation, and focus on vision as an example. I will provide some asymptotic theorems based on extensions of large-deviation theory. Through the asymptotic analysis, and our recognition experiments, I will argue that essentially optimal performance can be attained through a computationally feasible sequential decision analysis.