Abstract:
Many challenging pattern classification tasks require modeling of the complex stochastic processes using limited training data. In this talk we will present parsimonious and accurate modeling paradigms and discriminative model selection algorithms for such processes. This talk will consist of two parts. In the first part, we will introduce multi-rate hidden Markov models (multi-rate HMMs) as an extension of the widely used hidden Markov modeling framework, for multi-scale statistical modeling. The multi-rate HMM decomposes process variability into scale-based components and characterizes both the intra-scale temporal evolution of the process within each scale-based part and the inter-scale interactions among them. We have previously applied multi-rate HMMs to machine tool-wear monitoring problem, and in this talk we will present an application of the multi-rate HMMs to speech recognition, in which they are used for acoustic modeling of speech simultaneously at phone and syllable time scales. In the second part, we will introduce a discriminative model selection algorithm for statistical classifiers using the graphical modeling framework. The algorithm searches the space of graphical models for dependency structures that are inherently discriminative and gives practical and intuitive solutions to many problems encountered in classification system design, e.g. feature selection and dependency selection. We will use this model selection algorithm for extending prior work in multi-stream modeling by cross-stream observation dependencies and a new discriminative criteria for selecting such dependencies.
About the speaker:
Ozgur Cetin is a PhD student at the University of Washington Department of Electrical Engineering, where he is a member of the Signal, Speech and Language Interpretation Laboratory. He received his BS degree from Bilkent University, Turkey, in 1998 and his MS degree from University of Washington in 2000, both in electrical engineering.