Files in this item



application/pdfECE499-Sp2015-liu-Zuozhen.pdf (583kB)Restricted to U of Illinois
(no description provided)PDF


Title:A Study of Sampling-Based Hardware on Clustering Using Dirichlet Process Mixture Model
Author(s):Liu, Zuozhen
Contributor(s):Kumar, Rakesh
Dirichlet Process Mixture Model
Abstract:This thesis presents evaluation results on a hardware implementation of clustering using Dirichlet Process Mixture Model (DPMM). Clustering is a crucial unsupervised learning problem that has a wide range of applications in biology, medicine, social science, etc. The task of clustering is to classify a set of unlabeled data points into various clusters. DPMM is a sampling-based solution to clustering applications such as Neuron Spike Sorting in Lewicki (1998), where the number of clusters is an unknown parameter. We revised the algorithms and extended the preliminary hardware implementation in Verilog and software implementation in C from Ma (2014). We then conducted thorough performance evaluations and analysis by comparing results from the hardware implementation with the software. The simulation results of the Verilog implementation have shown very good performance in clustering synthesized data that is comparable to the software in various settings. The simulation results also helped understand the advantage and limitations of the hardware implementation, which lays a solid foundation for developing a general-purpose sampling-based hardware model in the future.
Issue Date:2015-05
Date Available in IDEALS:2015-08-03

This item appears in the following Collection(s)

Item Statistics