Learning on Distance Data with Applications in Cancer Research

• Applicant: Dr. Julia Vogt
• Subject Area: Theoretical Computer Science ; Bioinformatics and Theoretical Biology
• Term: from 2013 to 2016
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 238232249

Traditional machine learning methods usually depend on geometric information of the data. However, frequently no access is given to the underlying vectorial representation of the data but only pairwise similarities or distances are measured. Examples of data sets of this kind include all types of kernel matrices, be it string alignment kernels over DNA or protein sequences or diffusion kernels on graphs. The main goal of the project is to develop new machine learning methods based on relational data that do not require direct access to an underlying vector space. These new methods will then be applied to such data sets that have no obvious vector space representation. The application areas cover any data sets in form of pairwise distances. Since Mercer kernels can encode similarities between many different kinds of objects (for instance kernels on graphs, images, structures or strings) the methods proposed here will cover a broad scope of application. The main application area will be the analysis of cancer data provided by the Memorial Sloan-Kettering Cancer Center.

• DFG Programme: Research Fellowships
• International Connection: USA
• Host: Professor Dr. Gunnar Rätsch