check it out:
Thesis     -     Master Thesis

Programming Project    -    Werkcollege

Proposals 2009 - 2010

    1. Parallel Processing: Making parallel computational power used.
The ever-increase of the clock frequency of CPUs seems to have stopped. All big IT-companies, like Intel & Microsoft, have jumped on the parallel computing paradigm, since quadcores are becoming mainstream and octocores are next. Also graphics cards, containing up to 240 cores, are used more and more for high-performance computing. But the problem is to get the processing power used.
Parallel processing adds complexity to the applications which can only be handled succesfully by skilled parallel programming experts. This creates an enormous threshold towards mainstream usage of parallel computation. We are investigating several strategies to overcome these barriers, such as skeletal and generic programming, adding parallelism to domain-specific libraries, etc.
Dependending on the interests of the student, a more specific direction of his thesis topic will be defined.

Student Profile:

- C, C++ or java
- basic knowledge of software engineering
- follow the Parallel Systems course


    2. Machine Learning: Robust causal structure learning.
Causal structure learning is where machine learning and statistics meet. It is a challenging field and receives more and more attention from the industry, such as marketeers and the process industry. They are gathering statistical data more easily and want to extract knowledge from it. Learning the causal structure of a system from data is a type of data mining which relates closely to the theory of Bayesian networks. Algorithms developed by academics make learning possible, but suffer from too restricted assumptions which do not apply in real situations.
Recent insights in the matter allow for the construction of robust learning algorithms. The project will happen in close collaboration with the industry, which will provide real-world data.

Read and see more about it: => Research => Causal Inference, consult also the examples in the tutorial section.


Student Profile:

- knowledge of java
- knowledge of causal models and the learning algorithms is not required, they will be taught in the beginning of the project.

    3. Reconfigurable Hardware

FPGA's werden lang aanzien als a 'poor man's digital chip' (vs. zgn. ASICs).  Door hun fundamenteel ander economisch model (t.o.v. ASICs) blijken ze momenteel, tesamen met microprocessoren,  echter de meest ge-avanceerde chips te zijn.  Het doel van dit eindwerk is om 1 of 2 families van zgn. platform-FPGAs (i.e. een field programmable gate array, met on-chip meerdere von Neumann processoren) te bestuderen, te simuleren en te gebruiken voor een simulator, een grafische engine of een sorteerprocessor.
Een bijproduct van sommige Field Programmable Gate Arrays is dat ze her-programmeerbaar zijn.  Dit opent een interessant perspectief : wat gebeurt er als de chip (een deel van) zichzelf herprogrammeert als onderdeel van het algorithme ?  Het voordeel is evident : de beschikbare (hardware) resources kunnen beter benut worden dan in een ASIC of een Von Neumann computer.  Het probleem is ook evident : hoe moet men zoiets in software/firmware aanpakken?  Als resultaat van een doctoraat is er momenteel een prototype van een her-configureerbare computer beschikbaar.  Het doel van dit eindwerk is om de mogelijkheden en beperkingen van dit prototype (tesamen met de ontwerper ervan) te evalueren.
In English: Currently multi-Field Programmable Gate Array (FPGA) reconfigurable computing systems are still commonly used for accelerating algorithms. This technology where acceleration is achieved by spatial implementation of an algorithm in reconfigurable hardware has proven to be feasible. However, research pointed out that the application must fulfil some requirements in order to achieve a high performance. The best suiting algorithms are those.

    Read all about it ...: