My primary research area is design space exploration for algorithm/implementation co-optimization. Many applications in various disciplines are highly computationally-intensive (such as medical imaging algorithms). A direct implementation of these algorithms is either not possible (being too costly) or cannot meet desired requirements (performance/power). In such circumstances, the algorithm must be changed to meet requirements while the impact on application fidelity is kept to a minimum. We call it co-optimization of algorithm metrics (application fidelity) and implementation metrics (hardware resources, power requirements). It is not an easy goal to achieve due to the fact that algorithm and hardware designers speak in totally distinct languages. What we desperately need is a common platform for both parties to communicate changes that is intelligible to the other. We also develop tools that incorporate our latest research results and are deliverable to designers for field test.

Currently, we are conducting a project called ColSpace that rose from an ongoing biomedical image processing algorithm. It has real-time requirements and is highly data-intensive. We hope to take it as a case study and find a way to express algorithm and implementation collectively. The resulting tool shall be able to apply to a much broader range of applications.