August 30, 2012
The most common scenario in the machine learning setting is to represent data by a single vector, or graph space. However, in real life, multi-modal datasets exist. Each instance in these datasets has multiple representations (views) from various feature spaces. How to effectively combine all views in order to boost the efficacy of machine learning algorithms has led to the emergence of a new research area, called multi-view learning. This talk will focus on the clustering of multi-view data, providing useful insights into the problem and some novel multi-view approaches will be presented. These approaches address two prominent multi-view challenges, the diversity and the quality of the available views, from different perspectives. We will elaborate on the properties of our algorithms and how the above challenges are met, together with experimental results that support our claims.
Talk slides in pdf [~1MB]http://www.iit.demokritos.gr/docs/seminars/Tzortzis_Demokritos_presentation.pdf