Professor Tzi-cker Chiueh

Computer Science Building, Room 1419

chiueh AT cs DOT sunysb DOT edu (Email to make an appointment)

Tel: 632-8449

My belief is that every Master project should have a research and practical component to it. Therefore in each topic listed below, students are required to do a little research into the nature of the problem, and to carry out a solid implementation of the solution that arises out of the research. At the end of the project, the student is required to deliver a public presentation of the research results as well as a high-quality report. Those topics marked with "***" are co-sponsored by Dr. Paul Stirpe at Reuters.

The dominant load on future Web servers will be executing CGI programs rather than simply fetching files. Already this is the case with the advent of dynamic HTML. This project aims to implement a scalable Web server that can efficiently execute a large number of small CGI programs efficiently, using either threads or process pools. A detailed performance evaluation of the resulting prototype needs to be conducted after the implementation is in place.

Given a database of 3D objects, the goal of this project is to retrieve the most similar 3D object in the database to a given 3D object query. The basic approach is to define multiple object-specific projections for each 3D object, and compare and match them based on 2D projections. This project will be built on the 2D shape image indexing system developed in the PAMIS project.

The multi-dimensional nearest-neighbor indexing capability in PAMIS is generic enough that it can efficiently handle multimedia types whose distance metrics satisfy the triangular inequality. The goal of this project is to apply this indexing capability to the retrieval of speech documents. In particular, we are interested in exploring multi-resolution speech matching metrics that allows quick pruning of irrelevant can-didates in the early phase, and accurate pinpointing of target candidates in the later phase.

I/O plays an important role in efficient volume rendering applications. This project attempts to build a database system that provides effective I/O support to facilitate volume visualization applications. Specific techniques to explore in this project include compression, parallel I/O, sub-volume access, and 3D feature matching.

One of the important issues for modern superscalar machines is the management of on-chip memory, in particular, register files. In this project, we would like to develop new register allocation techniques for mul-tiple register files, each of which only has two read ports and one write port. We want to understand the per-formance difference between such architecture and truly single multi-ported register file, and how far intelli-gent register allocator can close this performance gap.