Projects

Holistic Pattern Recognition

Within this three year FFG Bridge7 project we research and develop intelligent image processing methods for the analysis of human tissue (placenta, colon and bone) using computerized slide-based microscopy. Intermediate results can be found in publications.

Handwritten Digit Recognition Revisited

I've revisited some assumptions about machine learning and found that, at least for the well-researched task of handwritten digits recognition, state-of-the-art machine learning systems are just as brittle as their old classical AI counterparts. Brittleness in this context means that their generalization performance on the whole task space (estimated by three distinct datasets) is very unsatisfactory -- they are unable to recognize handwritten digits in general, and the models are very specific to each dataset. You can find the empirically well-founded argumentation in this paper. This might be generally true as well, although that would be extremely hard to prove.

OCZ Neural Impulse Actuator - Driver for Linux

I have joined forces with Martin Bachem and we are working on a linux driver for the OCZ NIA, one of the hottest pieces of hardware in town. Progress is looking very good, but we could still use your help.

Internally, the hardware by Brain Actuated Technologies is used, which makes this a very good system at a very low price. We'd be happy to built custom applications - you just have to ask.

I have created a stand-alone autofocus application for Logitech USB cameras with proprietary motor focus extension. This application will allow you to automatically set the appropriate focus for a given scene. Download here: sw-af V0.05.

The underlying focus-from-image estimation has been integrated into guvcview, and beats all published approaches according to its developer. In fact, my algorithm will be the base of a unified software autofocus for all UVC compatible cameras via libv4l. Sadly, I don't have time to write a paper about this but maybe you have? I can offer guidance and test hardware.

Intelligent Go Board

In 2003, I have built a prototype 8x8 board which records Go moves during normal play, thus ensuring that a full game record is available for each game. No special stones are necessary, no holes need to be drilled through the board - it is a nice example for Ubiquituous/Pervasive Computing. Demonstration available upon request.

In 2006, I presented this project at the Idea Generation workshop, and four students created a business proposal for it.

2006/07 I also did some research on recovering the Go board state from still images with Image Mining and Machine Learning techniques. A technical report on this work can be found here.

2003-2005 I was involved in the BioMinT EU-funded research project, which was concerned with text mining in biological research literature and online databases. I was directly responsible for:

• Implementation of a parser for fourteen online databases, transitive closure on the link graph, output as name pairs (synonyms) - 12 million pairs in all.
• Research and development in biological species recognition, named entity recognition, ranking and filtering systems. Dataset
• Evaluation and validation of the system.
• Basic research on redundancy recognition (sentence entailment) Dataset

BlueCar

Within this project, I have modified a standard RC car for bluetooth remote control via a Symbian mobile phone. Could be combined with qik which allows video streaming directly from a mobile phone.