Implement and validate Affinity Propagation based hierarchical clustering on a GPU cluster

Status: Closed

Thesis: Master

Field: Statistics

Advisors: Oefner, Gronwald

Courses required: Practical Bioinformatics I&II

Objective: Affinity Propagation is a fairly new hierarchical clustering method that carriers over more conventional clustering techniques the advantage that decisions with regard to group affinity can be revisited. However, the method requires great amounts of Random Access Memory, which has limited the application of Affinity Propagation to relatively small datasets. Recently, Affinity Propagation has been implemented successfully on a Graphics Processing Unit and ongoing work aims at running Affinity Propagation on a GPU cluster to enable the clustering of up to one hundred thousand objects. The objective of this thesis is to apply GPU-based Affinity Propagation to various large datasets and to compare its performance to other clustering algorithms such as Ward's method.

Data: multi-color cancer cells, mass spectrometric and NMR fingerprints, job market data

First Steps: Familiarize yourself with unsupervised and supervised clustering methods, including Affinity Propagation, and GPU programming

Start reading: Frey BJ, Dueck D. Clustering by passing messages between data points. Science 2007, 315: 972–976.