Recent and ongoing research

Aspect-based Learning

Learning with an intermediate layer of feature predicates. Main advantages are a realistic model of relevance and similarity, simple and direct statistical analysis and very good scalability.

Collaborative Image Search Systems

Image retrieval systems based on relevance feedback allowing users to collaborate to improve their search results.

Visual Language Learning

Analysis of hierarchies of visual words. Working on visual phrases.

Dynamic Texture

Mining temporal texture patterns in video.

Nonlinear Statistical Inference for Large-scale Models

The topic of my PhD thesis. I am still fascinated by the question how well models can describe reality and how to measure this. The approach described in the thesis uses automatic differentiation of computer code to speed up the statistical computations.