Probabilistic Modelling

 
 
 
   

    

Postdoc: Max Welling 
  

Sponsors: Sloan Foundation 
 
Summary:  

The aim of this research is to develop probabilistic models for pattern recognition, efficient coding (data compression), denoising and automatic feature extraction. The probabilistic models under investigation are the following: 
 

  •  Linear Independent Component Analysis (ICA)
  •  Dryden-Mardia shape densities
  •  Non-linear graphical models (Baysian Belief Networks)
ICA searches for a basis in data-space onto which the data can be represented as independent as possible. In this sense it may be viewed as a generalization of PCA which decorrelates the data. 
An important application can be found in medical signal processing, where independent contributions to fMRI signals (e.g. heart beat, respiration, neural activity) can be separated with the use of ICA (see image below). 

 

   
  
 
 
 
 
 
Max Welling - welling@vision.caltech.edu