We are interested in how cell-signaling pathways pattern cell fates in developing tissues. The eggshell development of the fruit fly Drosophila melanogaster is an excellent model system to study how cell fates are determined. During egg development, the oocyte is surrounded by a layer of ~1000 uniform epithelial cells that differentiate into over 10 types of cells fates. This process relies on the non-uniform activation of two signaling pathways (EGFR and BMP). Many of the signaling components have been identified and a picture of the spatial distribution of signaling activity is emerging. Yet, a complete picture of what targets are downstream of these pathways and how they influence the development of the tissue remains. We use a range of genetic, genomic, and bioinformatic techniques aimed to provide a systems level understanding of how a tissue is patterned by multiple signaling pathways. For this we identify all of the targets of the EGFR and BMP pathways in the tissue with qRT-PCR and Affymetrix Gene Chips, characterize the spatial images of the gene expression patterns and archive them in a public database, and computationally identify and experimentally verify regulatory regions that mediate spatial-temporal patterns of gene expression. In addition, we explore the role of these newly identified targets in eggshell morphogenesis.

Emerging computer architectures have abandoned the traditional model of a single sequential stream of computation, in favor of one in which many independent computational units operate simultaneously, achieving dramatic increases in speed through parallelism. This is necessary in order to continue the rate of increase in computational performance that we have seen over the last few decades, as traditional architectures are no longer able to take advantage of the increasing capabilities of hardware production. However, this new poses additional challenges for computer scientists, as many algorithms that are efficient on one architecture are quite inefficient on the other. My research focuses on developing novel algorithms that utilize the potential of these new hardware architectures as applied to problems in scientific visualization and computer graphics. One major application is the extraction of surfaces from volume datasets, such as the surface of a liquid in a fluid simulation, or of organs in a MRI scan. Another application is the parallel processing of massive meshes generated from 3d scanning, such as in manufacturing, or for recording of cultural heritage artifacts, including paintings and sculptures. We have seen performance increases in these applications of orders of magnitude through the use of new parallel computation models, and we look to apply these to additional problems.

My dissertation work focuses on the development and exploration of novel approaches to embedded systems for computer vision and media processing. In addition, I consider optimized implementations of computer vision applications mapped to parallel computing platforms. My consideration of embedded systems is done with a widely cast net. On one end of the spectrum, I consider conventional embedded systems, such as those found in portable electronic devices. As consumer-drive performance demands increase, and process technology scaling ceases or is unable to provide means for meeting these demands, these conventional systems have evolved into Multiprocessor Systems on Chip (MpSoC), with expected parallel computing attributes and considerations. On the other end of the spectrum, I consider conventional high performance computing machines, such as supercomputing clusters. It is my opinion and experience that these machines, from a pragmatic perspective, tend to be tailored towards application-specific implementation, thus fitting the accepted definition of an embedded system. As research pushes past conventional architectures, heterogeneous designs are paramount. Accordingly, my work also considers points between the two aforementioned ends of the spectrum. I have looked at heterogeneous or hybrid embedded vision systems consisting of specialized components. These include FPGA accelerators, GPU-accelerated processing nodes within clusters, commercial off-the-shelf (COTS) processors such as the IBM Cell MpSoC and Texas Instrument's DaVinci DSP. Furthermore, I am in the process of developing a hybrid accelerator consisting of two of these specialized components. My overall dissertation goal is to provide generalized architectural guidelines for embedded vision system designers.