Date: Oct 1st, 2010
Title: Automatic Detection of Sub-Kilometers Craters in High Resolution Planetary Images
Abstract: Due to non-distinguishing features and the heterogeneous morphology, the discovery of sub-kilometer craters in high resolution planetary images is a most challenging problem. Counting craters in remotely sensed images is the only tool that provides relative dating of those remote planetary surfaces. Accurately dating surface in high spatial resolution requires counting a large amount of small sub-kilometer craters, which calls for highly efficient automatic crater detection. In this paper, we present an integrated framework on auto-detection of sub-kilometer craters. The framework uses an innovative method that contains three key components. First, we utilize mathematical morphology to efficiently identify crater candidates, the regions of an image that can potentially contain craters. Only those regions, occupying relatively small portions of the original image, are the subjects of further processing. Second, we extract and select image texture features, in combination with supervised boosting ensemble learning algorithms, to accurately classify crater candidates into craters and non-craters. Third, we integrate transfer learning into boosting, to enable maintaining an accurate auto-detection of craters on new images. Our framework is evaluated on a large test image of 37,500 meters by 56,250 meters on Mars, which exhibits a heavily cratered Martian terrain characterized by nonuniform surface morphology. Our transfer learning algorithms can enhance detection performance in the regions where surface morphology differs as characterized by the training set. Empirical studies demonstrate that the proposed crater detection framework can achieve an F1 score above 0.85 while existing algorithms can only be slightly better than 0.4 at best.
Speaker's short bio: Wei Ding has been an Assistant Professor of Computer Science at the University of Massachusetts Boston since 2008. She received her Ph.D. degree in Computer Science from the University of Houston in 2008. Her main research interests include Data Mining, Machine Learning, Artificial Intelligence, Computational Semantics, and with applications to astronomy, geosciences, and environmental sciences. She has published more than 30 referred research papers and has 1 patent. She is the recipient of a Best Paper Award at IEEE ICCI 2010, a Best Poster Presentation award at ACM SIGSPAITAL GIS 2008, and the Best PhD Work Award between 2007 and 2010 from the University of Houston. Her research projects are currently sponsored by NASA, DOE, and NSF.
Date: Oct 8th, 2010
Title: Malicious Shellcode Detection with Virtual Memory Snapshots
Abstract: Shellcodes are short sequences of instructions injected into and then executed by an exploited process. Malicious shellcode injection is a widely adopted cyber attack technique. It is a key technique used throughout the entire history of worm attacks from the Morris worm in 1988, which infected 10% of all computers connected to the Internet at that time, to today’s Conficker worm, which has infected millions of computers worldwide. Malicious shellcode injection also accounts for most system break-ins and subsequent disabling of security measures, opening of backdoors that grant unauthorized remote access to systems, and downloading or activation of further malicious codes. As shellcode detection technology becomes more prevalent, shellcode continues to evolve. State-of-the-art shellcodes can exploit runtime information of a target process and use techniques such as polymorphism and metamorphism to evade detection. Timely detection of such shellcodes is challenging. In this talk, we present a novel malicious shellcode detection system. Facilitated by virtual memory snapshots that capture runtime information of a target process, our system detects malicious shellcodes by revealing their real behavior. Our system can be used to effectively defend against zero-day malicious shellcode injection attacks. We implemented the system in Debian Linux with kernel version 2.6.26. Our experiments with extensive real traces and thousands of malicious shellcode instances validate our system’s performance with low overhead and few false negatives and positives. In addition, this talk briefly presents our other work on security and future research directions.
Xiaole Bai is an assistant professor in computer and information science department. He received his B.S. degree in 1999 at Southeast University, China, and a M.S. degree in 2003 from the Networking Laboratory at Helsinki University of Technology, Finland. He joined The Ohio State University in 2004 and received his Ph.D. degree in December 2009 from the Department of Computer Science and Engineering. Please refer to his webpage: http://www.cis.umassd.edu/~xbai/ for more information.
Date: Oct 29th, 2010
Title: Access control and distributed resources assignment in optical burst-switched networks
Abstract: The emergence of a broad range of network-driven applications (e.g., multimedia, online gaming) brings in the need for a network environment able to provide multiservice capabilities with diverse quality-of-service (QoS) guarantees. In this
talk, a medium access control protocol is introduced to support multiple services and QoS levels in optical burst-switched mesh networks without wavelength conversion. The protocol provides two different access mechanisms, queue-arbitrated and prearbitrated for connectionless and connection-oriented burst transport, respectively. It has been evaluated through extensive simulations and its simplistic form makes it very promising for implementation and deployment. The overall results demonstrate the suitability of this architecture for future integrated multiservice optical networks.
Speaker Bio: Joan Triay received a B.Eng. and a M.Eng. in telecommunications engineering and a M.Sc. in telematics engineering, in 2004, 2006, and 2007, respectively, all at Universitat Politècnica de Catalunya (UPC), Spain. In 2007, he was awarded with a 4-year predoctoral FI scholarship from the Government of Catalonia and the European Social Fund, and since then he is a Ph.D. candidate in the Department of Telematics Engineering at UPC. He was a visiting fellow at the University of Essex (UK) from June 2009 to August 2010 thanks to a BE-DGR fellowship. Currently, he is a visiting researcher at University of Massachusetts, Dartmouth, with the support of a Fulbright graduate fellowship. His research interests include, but are not limited to, future optical network architectures, optical control plane design and the provisioning of multiservice capabilities on high-speed networks.
Date: Nov. 19th, 2010
Title: Information Brokerage in Sensor Networks
Abstract: Today's sensor networks are deployed primarily for
collecting sensor data from the field to a central sink for analysis.
As sensor technologies continue to improve, resulting in low-cost
sensor platforms with increased capacities, more and more sensor
networks will be enabled to serve sophisticated purposes, with not
only the sensing and collecting functions but also the capability to
store and process information on the fly inside the network. Essential
to the principled design of such networks is the problem of
information brokerage: how the nodes coordinate with each other so
that any information, when generated or captured at a node, can be
brought quickly and efficiently to the attention of those nodes
interested in this information. An adequate solution to this problem
will enable many important applications and services of sensor
networks. For example, in industrial settings, we can provide a more
secure working environment by deploying a sensor network that warns
workers upon detection of dangerous events; in this application, there
need to be underlying efficient mechanisms to quickly inform the
workers when such an event occurs. Group-communication services can
also be set up easily to allow any number of nodes to receive the same
data multicast from one or more sources in the network. This talk is
focused on the challenges as well as the current techniques for
information brokerage in sensor networks in support of both
topic-based and content-based services.
Duc A. Tran is an Assistant Professor in the Department of
Computer Science at the University of Massachusetts at Boston, where
he leads the Network Information Systems Laboratory (NISLab). He
received a PhD degree in Computer Science from the University of
Central Florida (Orlando, Florida) in 2003. Dr. Tran's research
interests are focused on data management and networking designs for
decentralized networks such as P2P networks, sensor networks, and
disruption-tolerant networks. The results of his work have led to
research grants from the National Science Foundation (NSF), a Best
Paper Award at ICCCN 2008, and a Best Paper Recognition at DaWak 1999.
Dr. Tran has engaged in many professional activities, serving as a
Review Panelist for the NSF, Editorial Board member for the Journal on
Parallel, Emergent, and Distributed Systems (2010-date) and ISRN
Communications Journal (2010-date), Guest-Editor for the Journal on
Pervasive Computing and Communications (2009), TPC Co-Chair for CCNet
2010, GridPeer (2009, 2010, 2011), and IRSN 2009, and TPC Vice-Chair
for AINA 2007.