Research Fellow  |  Lee, PeiZong  

        PeiZong Lee received a B.S. degree in mathematics from National Taiwan University in 1979, an M.S. degree in computer and decision sciences from National Tsing Hua University, Taiwan, in 1984, and a Ph.D. degree in computer science from the Courant Institute of Mathematical Sciences, New York University, in 1989. Between 1984 and 1986, he was a research engineer at the Institute of Information Industry, Taiwan, where he participated in an artificial intelligence project. He joined with the Institute of Information Science, Academia Sinica in 1989, currently, he is a Research Fellow.

        His research interests are in parallel algorithm design, compilers for scientific applications, and digital signal processing. He now also participates in a numerical wind tunnel project, in which he and other colleagues have to conquer researches for unstructured mesh generation, Euler equation and Navier-Stokes equation solvers, visualization, and the challenge of using MPI on a workstation cluster to accelerate the irregular computation.

        He has five major contributions for compiler research: First, for constant data-dependence applications, he found necessary and sufficient conditions for mapping nested For-Loop algorithms into systolic arrays.  Second, for data dependences known at compiling time, he proposed a dynamic programming algorithm to arrange data distribution for each data array on distributed memory parallel computers.  Third, also for applications whose data dependences known at compiling time, he further ranked dominant data arrays whose migration would be the most expensive, so he could proposed algorithms to decompose data and computation on distributed memory parallel computers simultaneously. Fourth, he and other colleagues integrated inspector-executor techniques to handle compiler and run-time support for irregular programs, whose data dependences not known at compiling time but only known until at run-time.  Fifth, since automatic compilation for irregular applications is still an open problem, for a large class of applications based on unstructured meshes, whose data array indices are other array elements, therefore, data dependences are not known at compiling time, he has proposed a paradigm for designing parallel CFD programs based on domain decomposition techniques.