Research Buffet
Feifei Li
University of Utah
Wednesday, September 12, 2012
3147 MEB
Lecture 3:40 p.m.
Title: Towards Building Local Search Engines over Large Heterogeneous Data
Feifei Li
University of Utah
Wednesday, September 12, 2012
3147 MEB
Lecture 3:40 p.m.
Title: Towards Building Local Search Engines over Large Heterogeneous Data
Abstract
In many applications over large heterogeneous data, search is still executed via formbased query interfaces, which are then translated into SQL statements to find matching records. Ranking is usually not implemented unless users have explicitly indicated how to rank the matching records, e.g., in the ascending order of year. Often, this approach is neither intuitive nor user-friendly (especially with many search fields in a query form). It also requires application developers to design schema-specific query forms and develop specific programs that understand these forms. In this work, we propose to demonstrate the ColumbuScout system that aims at quickly building and deploying a local search engine over one or more large databases. The ColumbuScout system adopts a search-engine-style approach for searches over local databases. It introduces its own indexing structures and storage designs, to improve its overall efficiency and scalability. We will demonstrate that it is simple for application developers to deploy ColumbuScout over any databases, and ColumbuScout is able to support search-engine-like types of search over large databases efficiently and effectively.
Cem Yuksel
University of Utah
Title: Developing Efficient Graphics Technologies for Physically-based Modeling
In many applications over large heterogeneous data, search is still executed via formbased query interfaces, which are then translated into SQL statements to find matching records. Ranking is usually not implemented unless users have explicitly indicated how to rank the matching records, e.g., in the ascending order of year. Often, this approach is neither intuitive nor user-friendly (especially with many search fields in a query form). It also requires application developers to design schema-specific query forms and develop specific programs that understand these forms. In this work, we propose to demonstrate the ColumbuScout system that aims at quickly building and deploying a local search engine over one or more large databases. The ColumbuScout system adopts a search-engine-style approach for searches over local databases. It introduces its own indexing structures and storage designs, to improve its overall efficiency and scalability. We will demonstrate that it is simple for application developers to deploy ColumbuScout over any databases, and ColumbuScout is able to support search-engine-like types of search over large databases efficiently and effectively.
Cem Yuksel
University of Utah
Title: Developing Efficient Graphics Technologies for Physically-based Modeling
Abstract
As we try to build virtual representations for more and more complicated environments in graphics, computational efficiency often becomes the determining factor in how useful a particular graphics technique is in practice. In this talk I will present my approach for developing efficient graphics technologies, which consists of three steps: visual analysis of the subject physical phenomenon, understanding of the physical models that describe the phenomenon, and devising algorithms that fit well to the target computer hardware. The visual analysis helps separate the phenomenon into its visual components and identify which one of these components have deterministic behavior, while understanding the physical models makes it possible to propose simpler models that properly approximate the important visual components. Finally, efficient algorithms can be developed for simplified physical models considering the architectural strengths of the desired computer hardware.
I will explain how this approach helped in developing efficient technologies by presenting two separate research topics as examples: water simulation and realistic hair rendering. For water simulation I will talk about the wave particles method for real-time simulation of water surface waves. The wave particles method can achieve about 6 orders of magnitude speed up as compared to the most efficient full fluid simulations in graphics. As a result, scenes with very high complexity can be simulated with high frame rates. As for hair rendering, I will present the dual scattering method for approximating multiple scattering of light in human hair. Multiple scattering plays a crucial role in defining the perceived color of hair and a physically based computation of multiple scattering is important for generating realistic hair images. While previous methods required hours to compute multiple scattering for a single hair image, the dual scattering method can properly approximate the same computation at real-time frame rates with high visual accuracy.
As we try to build virtual representations for more and more complicated environments in graphics, computational efficiency often becomes the determining factor in how useful a particular graphics technique is in practice. In this talk I will present my approach for developing efficient graphics technologies, which consists of three steps: visual analysis of the subject physical phenomenon, understanding of the physical models that describe the phenomenon, and devising algorithms that fit well to the target computer hardware. The visual analysis helps separate the phenomenon into its visual components and identify which one of these components have deterministic behavior, while understanding the physical models makes it possible to propose simpler models that properly approximate the important visual components. Finally, efficient algorithms can be developed for simplified physical models considering the architectural strengths of the desired computer hardware.
I will explain how this approach helped in developing efficient technologies by presenting two separate research topics as examples: water simulation and realistic hair rendering. For water simulation I will talk about the wave particles method for real-time simulation of water surface waves. The wave particles method can achieve about 6 orders of magnitude speed up as compared to the most efficient full fluid simulations in graphics. As a result, scenes with very high complexity can be simulated with high frame rates. As for hair rendering, I will present the dual scattering method for approximating multiple scattering of light in human hair. Multiple scattering plays a crucial role in defining the perceived color of hair and a physically based computation of multiple scattering is important for generating realistic hair images. While previous methods required hours to compute multiple scattering for a single hair image, the dual scattering method can properly approximate the same computation at real-time frame rates with high visual accuracy.