My most recent project involved developing an algorithm that can classify 3D confocal microscope images of stem cells as having been grown on a medium that induces differentiation or on a medium that does not induce differentiation. The algorithm is easily parallelizable and I have developed both a serial CPU implementation as well as several parallel GPU implementations based on either atomic operations or a parallel reduction operation.
Another recent project was developing a parallel, GPU-accelerated algorithm to measure the surface area of molecules whose atoms are modelled as Gaussian Radial Basis Functions (RBFs). In addition to being parallel the algorithm is also progressive, returning a rough answer almost immediately and then refining that answer over time to the desired level of accuracy.
Some of my other projects in this area have involved fitting 3D medical and physics data with Gaussian Radial Basis Functions (RBFs) and developing algorithms to render the data directly from this representation, and investigating the segmentation of gene expression data using spectral algorithms which use eigenvectors of Laplacian matrices derived from the data.
I have also assisted in the design and construction of a 24 node Linux visualization/computation cluster. The cluster nodes are equipped with dual CPUs and high-end GPUs, and are connected to a 5x5 tiled LCD display wall with a resolution of 100 megapixels. This project is being funded by the NSF to investigate the benefits of using graphics hardware's parallel architecture in conjunction with standard CPUs in a cluster computing environment.