6:00 - 6:15pm: Luke and Tomasz to open diet Pepsi and cookies, introduction of new people,
6:15 - 6:35pm: Mark Ho "Intersection Marker Method for tracking a deformable 2D surface in 3D Eulerian Space",
6:35 - 6:55pm: Pawel Lachowicz "GPUs in Modern Quantitative Finance",
6:55 - 7:15pm: Michael Wilkinson "Satellite Data Analysis using Parallel Computation of Differential Morphological Profiles",
7:15 - 7:25: Tomasz Bednarz "Updates on newest GPU based demoscene productions",
7:25 - 9:00pm: dinner at MQ Shopping Centre.
Mark Ho "Intersection Marker Method for tracking a deformable 2D surface in 3D Eulerian Space": A new method for tracking a deformable interface in 3D Eulerian space has been developed. The method can model an arbitrary 3D shape immersed inside an array of uniform hexahedral control volumes by using a combination of planar polygons. Since each planar polygon intersects the edges of the control volume and the combination of cell-edge intersections uniquely identifies the type of polygon a control volume holds, this new explicit tracking method has been named the ‘(I)nter(S)ection (M)arker’ (ISM) method for interface tracking. The intended application of the ISM method is in CFD simulations of multiphase flows.
Pawel Lachowicz "GPUs in Modern Quantitative Finance": Today's finance requires the enormous computer speed for recalculation of market positions, hedging own investments, estimating risk. Recent GPU solutions employing CUDA architecture seem to bridge the gap between the usage of multi-core CPU clusters and super-computers. At effectively lower cost, they provide a noticeable acceleration in computations. I will briefly highlight a current software development status and trends in the quantitative finance and risk management making use of NVIDIA CUDA.
Michael Wilkinson "Satellite Data Analysis using Parallel Computation of Differential Morphological Profiles" When analysing satellite images to detect rubble and other damage after disasters such as earthquakes, Differential Morphological Profiles (DMP) are a powerful tool. However, classical algorithms would need 34,000 years on a single CPU, for the 1.5 terapixels of the Haiti earthquake data set. Faster algorithms cut this down to 40 days, but this is still too slow. A new, concurrent, shared memory parallel algorithm allows processing of 1.2 gigapixel per second on a 24 core machine. For tile-wise processing of the enite data set on a 16 node cluster this yields 1.5 hours of processing time.
Tomasz Bednarz "Updates on newest GPU based demoscene productions" If time allows, 2-3 newest demoscene productions will be shown :-)