Date of Award
Spring 5-2018
Degree Name
Master of Science (MS)
Degree Type
Thesis
Abstract
Pattern recognition is widely used in many areas. When analyzing a pattern recognition system, one of the main problem to be considered is, how many bits do I need to express the raw source data and memory data to ensure that the result of the pattern recognition be reliable. The data stored in the system as well as the data received by the system must be compressed by some rate to summary the raw data. The fundamental bound for this lies in the computation of the achievable rate of pattern recognition. Before now, we have the definition and some approaches for this achievable rate region from an information theory point of view, but these approaches can be applied only to some specific cases. There’s need for a method to compute this region’s boundary and this method should be able to be extended to any general case.
In this thesis, we present a new optimization algorithm associated with other algorithms in alternating optimization problems. This new algorithm will compute a bound of the achievable rate region in pattern recognition by solving the associated optimization problem. We show that this new algorithm can solve the problem we have for computing the boundary of the achievable rate region and can be extended to other areas.
Language
English (en)
Chair
Joseph O'Sullivan Zachary Feinstein Hiro Mukai
Committee Members
Joseph O'Sullivan Zachary Feinstein Hiro Mukai
Comments
Permanent URL: https://doi.org/10.7936/K74F1Q4T