Parallel Nonnegative Matrix Factorization for Data Analysis with CUDA
Organization
Algorithms & Theory
Abstract
Nonnegative Matrix Factorization (NMF) is a way to automatically decompose frequencies (number of times a word occurs on a web page) into a small number of basic frequency patterns (corresponding to topics). Computing an NMF with smallest error is a difficult non-convex optimization problems.
There are several algorithms which perform NMF. The goal of this thesis is to develop one which can be effectively and efficiently parallelized for execution on GPUs.
The resulting code will use CUDA.
There are several algorithms which perform NMF. The goal of this thesis is to develop one which can be effectively and efficiently parallelized for execution on GPUs.
The resulting code will use CUDA.
Graduation Theses defence year
2016-2017
Supervisor
Dirk Oliver Theis
Spoken language (s)
English
Requirements for candidates
C or C++. Interest in parallel programming, GPU, CUDA.
Level
Bachelor, Masters
Keywords
Application of contact
Name
Dirk Oliver Theis
Phone
E-mail