Colloquium, "Towards Building a Self_learning Environment", Dr. Vijay V. Raghavan

Abstract

EThe main goal of this presentation is to introduce the need for developing the Semantic Web and overview existing technologies to enable its creation. Since much of the information on the Internet is semi-structured or unstructured and is primarily structured for use by humans, it is highly challenging to build automated, intelligent applications that can use that information in the same way humans can. One of the main challenges is the creation of knowledge bases that can aid machines to correctly interpret the vast amounts of textual data on the Internet. We then introduce the two major alternative approaches to achieve this, namely: i) Natural Language Processing, and ii) Semantic Web.

Following the introduction, a detailed discussion of the Semantic Web technologies and standards, includ- ing the Layer-cake Architecture for the Semantic Web is provided. In our research group at CACS, we currently have a research project aimed at developing a software framework that will be beneficial for simplifying and speeding up the process of building semantic applications. An overview of that frame- work, along with a case study that illustrates our ideas on Wikipedia-based approach to bootstrapping the knowledge base generation, is presented next. Then, we briefly review a few important semantic web tools. Finally the conclusions of the case study and directions for future work are presented.ssential for the Internet in handling its massive data volumes daily, routers connect wires operating at various speeds, possibly up to 40 Gbps. Fast IP address lookups have become indispensable for all routers, whether core or edge ones, but they are challenging, involving longest prefix matching (LPM, which chooses the longest prefix among those which match the given IP address). Earlier solutions for IP address lookups were mostly trie-based to match an IP address progressively a few bits at a time against prefixes stored in a tree-like data structure. Given rapid surges in the routing table sizes and widespread deployment of 10Gbps Ethernet switches lately, the LPM lookup rate of tens of million of packets per second (MPPS) cannot be achieved by classical LPM solutions, prompting active research. This talk sur- veys representative trie-based LPM methods, examines hash-based LPM techniques, and briefly intro- duces our proposed LPM scheme with superior storage-efficiency to yield over 100 MPPS for high-per- formance routers.

DATE: FRIDAY, FEBRUARY 5, 2010
TIME: 11:00 A.M. - 12:00 NOON
LOCATION: ACTR AUDITORIUM, ROOM 112