M4 – Machine Learning for Information Management & Analysis

advertisement
N1 − Technical Session: Network Planning, Design, and Operation
Session Chair
Ben Tang
600-700 Mountain Avenue
Murray Hill, NJ 07974
btang@alcatel-lucent.com
BIAOGRAPHY
Dr. Ben Tang is a distinguished member of technical staff in the Network Modeling and Optimization
Group at Bell Laboratories. He joined AT&T Bell Labs in 1991, working on software development for
5ESS and advanced decision support system. Since 1996, he has been working on the planning, modeling
and design of end-to-end network solutions.
Currently, Dr. Tang’s work focuses on all aspects of data networking, including architecture, traffic
modeling, network design and optimization, evolution planning and economic analysis, as well as modeling
and end-to-end solution development for emerging topics in next generation networks, such as IMS, triple
play broadband access, IPv6, content delivery and LTE. His work also includes the development of
advanced IP/MPLS network design methods and tools. Dr. Tang has led the work on numerous global
service provider projects to enhance their network efficiency and performance and reduce total cost of
ownership.
Dr. Tang has a B.S. degree from the National Taiwan University, M.S. from the University of Florida, and
Ph.D. from Purdue University, all in electrical engineering. He was invited as a member of
Telecommunication Advisory Board for the Ministry of Transportation and Communications, Republic of
China, in 1999. He was the session chairman at several telecommunications conferences.
N1 − Technical Session: Network Planning, Design, and Operation
Additional Switching Nodes: Not a Panacea
for Congested Wireless Networks
Amit Mukhopadhyay, John Zhao
Bell Laboratories, Alcatel-Lucent
600-700 Mountain Avenue, Murray Hill, NJ 07974
amitm@alcatel-lucent.com
zjzhao@alcatel-lucent.com
ABSTRACT:
Traditionally, service providers have been adding Mobile Switching Centers (MSCs) in their wireless networks
whenever the switches run out of capacity. In a recent analysis of a large metropolitan network, we observed
something seemingly counter-intuitive: additional MSCs do not necessarily add subscriber capacity and they can
sometimes even cause decreased network capacity.
The apparent anomaly can be intuitively explained by noting that the subscriber capacity of a network is a function
of both its call-processing capacity as well as mobility-handling capacity. A new MSC in a network will always add
call-processing capacity but additional MSCs will also result in increased inter-MSC signaling for mobility
handling. Exhaustion of signaling capacity may offset any benefit of the additional call-processing capacity of the
new MSC.
In this paper, we establish a methodology for predicting capacity exhaust in an expanding network and also present
practical suggestions for avoiding network congestion. Even though the analysis was done for a network with
traditional monolithic switches, the methodology can be applied towards next generation distributed network
elements as well.
BIAOGRAPHY:
Amit Mukhopadhyay is a Distinguished Member of Technical Staff in the Network Planning, Performance and
Economic Analysis Center in Bell Labs, New Jersey. His current work focuses on next generation wireless
technologies, including access and core networks for LTE/ePC and WiMAX. His networking interest spans other
wireless technologies, e.g., GSM/GPRS/EDGE, UMTS/HSPA, CDMA2000-1x/EVDO/UMB, DVB-H/MediaFlo
etc. He is also deeply involved in converged IMS networks with other broadband access technologies including
DSL, HFC Cable and Fiber. He has received numerous internal awards and is a member of the Alcatel-Lucent
Technical Academy.
He holds a B.Tech. in Naval Architecture from Indian Institute of Technology, Kharagpur, India and a Ph. D. in
Operations Research from the University of Texas, Dallas. He is a Senior Member of the IEEE, serves as an officer
in the AP/EMC/VT Chapter in IEEE NJ Coast Section and received the IEEE Region 1 award. He has numerous
publications in refereed journals and has received one patent award.
John Zhao is a Member of Technical Staff in the Network Modeling & Optimization Group in Bell Labs, New
Jersey. His current work focuses on design and deployment of scalable multi-service networks with MPLS/IP, metro
Ethernet, broadband access, and their inter-working technologies. His recent patent (pending) describes the
optimized design of implementable resilient Ethernet. He is also interested in developing innovative planning and
design methods and procedures to optimize both core and access network architectures in NGN deployment.
John Zhao received his B.Sc. degree in Automatic Control Engineering from Nanjing Institute of Technology in
China and his Ph.D. degree in Electrical Engineering from Polytechnic University, Brooklyn, New York. He is an
active member of IEEE Communication Society and a member of ACM/SIGCOMM.
N1 − Technical Session: Network Planning, Design, and Operation
Packet Optical Transport Network Architecture Impact on
Carrier Migration Strategy
Mohcene Mezhoudi, Poudyal Vijaya
Alcatel-Lucent Technologies
mezhoudi@alcatel-lucent.com
ABSTRACT: Carriers are decreasing their dependence on SONET/SDH and ATM as they move from TDM and
single wavelength transmission to packet and multi-wavelength transmission. Market research predicts that in 2009
there will be less spending on SONET/SDH than on WDM in North America. It is also forecasted that worldwide
WDM spending will overtake SONET/SDH in 2010. A packet optical transport network carries packet traffic,
typically on Ethernet interfaces, on SONET/SDH and/or WDM gear. Surveys show that the majority of Carriers
have packet optical transport network now and almost 70% will have it by the end of 2009. It is the general industry
belief that packet optical transport networks bring efficiencies and the majority of carriers expect operational
expenditure (OPEX) savings from packet transport.
The burning question that needs to be discussed is what is the correct migration strategy that will allow Carriers to
realize these efficiencies? Many carriers are taking a dual approach: using Packet/SONET/SDH/WDM optical
transports equipment in parts of their networks, and going straight to Packet/WDM/ROADM products in other parts.
In addition, many carriers operate multiple backbone networks (SONET/SDH, WDM, IP/MPLS) separately which
complicates the migration strategy further since there is a need to converge multiple networks infrastructures in
order to achieve the anticipated CAPEX/OPEX savings. In this talk we identify various approaches to the
architectural solutions to the packet optical transport (POT) network. A hierarchical architecture uses a converged
routing/switching platform at the core, with flexible OTN platform (ROADM) at the Core and Edge. A different
approach to packet optical transport architecture will be based on flexible OTN platforms to interconnect both Core
and Edge nodes, and leaves the switching to the IP/MPLS layer. The POT architecture choice, together with the
introduction of ultra-high speed transmission such as 40G/100G and next-generation multi-degree ROADM with
high channel density (80 Channel or higher) will have a very strong impact on the Carrier migration strategy. It
would involve the economical phase-out of existing platforms and a thorough business case study of each approach
is required. It is recognizable that the migration phase of existing customers to the new network infrastructure will
be costly in terms of operational expenditure in the short term but the overall payback period will depend on the
existing network and selected migration plan and needs a case by case study.
BIOGRAPHY:
Dr. M. Mezhoudi got his M.S.E.E. and Ph.D. From Stevens Institute of Technology, New Jersey. After teaching at
Stevens as an assistant professor and running the Optical Communications laboratory at Stevens, he joined Bell
Laboratories as a Member of Technical Staff in 1995. In 1999, he became a Distinguished Member of Technical
Staff. He then was promoted to Consultant Member of Technical Staff.
Dr. Mezhoudi has several publications in technical journals and conferences. He was awarded twice the Bell Lab
President award. His current research involves optical network transport, packet/optical network switching and
routing optimization techniques and reliability
Dr. V. Poudyal received the M.E. and Ph.D. degrees in electrical engineering from Stevens Institute of Technology,
Hoboken, NJ. He was formerly a Senior Systems Engineer at Telcordia Technologies, Inc., Piscataway, NJ and a
Lecturer of electrical engineering at the Institute of Engineering, Kathmandu, Nepal. He has advised local
government agencies on various communications technology issues as a private consultant. He is currently a Senior
Systems Engineer working for the Optical Networking Group of Alcatel-Lucent. For the past 10 years, he has been
working on various aspects of optical networking, including SONET and DWDM network design, DWDM
equipment specifications and standards, metropolitan and ultra long haul network design, and network design
software tools specification. His current research is on multilayer and multi-period network optimization
techniques and tools.
N1 − Technical Session: Network Planning, Design, and Operation
An IPv6 Migration Economic Study
Ben Tang
Bell Laboratories, Alcatel-Lucent
600-700 Mountain Avenue, Murray Hill, NJ 07974
btang@alcatel-lucent.com
ABSTRACT:
The global pool of public IPv4 addresses is quickly running out. Service providers, driven by the exhaustion of IPv4
addresses along with growing IP endpoints (in wireless and sensor network, for example) and the need for a public
IP address for emerging applications such as those based on peer-to-peer, are faced with the challenge of migrating
to IPv6. This study compares two migration scenarios for a fixed residential broadband access service provider - one
scenario which continues to use private IPv4 addresses after the exhaustion of public IPv4 addresses and deploy
Service Provider NAT (SP-NAT) for connection to the global Internet, and another scenario which introduces IPv6
immediately after the address exhaustion through the deployment of Dual Stack (DS) residential gateways. The
study examined the consequent capital investment and operations expense incurred over a multi-year period in the
two migration scenarios based on possible penetrations of DS residential gateways, DS hosts, global IPv6 endpoints
and IPv6 supporting applications. The impact of an industry formed IPv6 Consortium on the costs of the migration
scenarios was also addressed. Sensitivity analysis was performed to identify most influential factors affecting the
comparison of the two migration scenarios. The results of the study help answer key questions for service providers,
such as which migration scenario is more cost effective, what is the best timing to introduce IPv6, how does the
IPv6 Consortium affect the economics of IPv6.
BIAOGRAPHY:
Dr. Ben Tang is a distinguished member of technical staff in the Network Modeling and Optimization Group at Bell
Laboratories. He joined AT&T Bell Labs in 1991, working on software development for 5ESS and advanced
decision support system. Since 1996, he has been working on the planning, modeling and design of end-to-end
network solutions.
Currently, Dr. Tang’s work focuses on all aspects of data networking, including architecture, traffic modeling,
network design and optimization, evolution planning and economic analysis, as well as modeling and end-to-end
solution development for emerging topics in next generation networks, such as IMS, triple play broadband access,
IPv6, content delivery and LTE. His work also includes the development of advanced IP/MPLS network design
methods and tools. Dr. Tang has led the work on numerous global service provider projects to enhance their network
efficiency and performance and reduce total cost of ownership.
Dr. Tang has a B.S. degree from the National Taiwan University, M.S. from the University of Florida, and Ph.D.
from Purdue University, all in electrical engineering. He was invited as a member of Telecommunication Advisory
Board for the Ministry of Transportation and Communications, Republic of China, in 1999. He was the session
chairman at several telecommunications conferences.
N1 − Technical Session: Network Planning, Design, and Operation
Network Optimization Tools for Complex Networks
Ming-Jye Sheng
The MITRE Corporation, Eatontown, NJ, USA
msheng@mitre.org
Thomas Mak
US Army, PM WIN-T, Fort Monmouth, NJ, USA
ABSTRACT:
The algorithms/models proposed in the literature for TCP/IP networks, Mobile Ad-hoc Networks (MANETs), and
Sensor networks are quite complex. Currently much of the evaluation of these network mechanisms has been done
using high fidelity discrete-event simulation tools like OPNET Modeler. Development of models for these complex
protocols is time consuming and expensive. An analytical approach to the simulation results could provide a better
test coverage for network plan and design.
The method is based on the construction of a mathematical model of the algorithm/protocol of interest by representing
all the possible states of the algorithm/protocol, and the probabilities of the transitions that can occur between these
states. We develop formal techniques and tools to systematically and accurately assist with the design, testing and
performance analysis of MANETs by specifying states and transition probabilities of models and evaluating the
satisfaction probabilities over the state space; furthermore, approximate algorithms and statistical methods are
developed to overcome state explosion problems caused by larger and more complex models of the networks of
interest. This approach will correlate these models with simulation results and real world performance results to
benefit network deployment.
BIAOGRAPHY:
Ming-Jye Sheng, is currently a member of MITRE’s network, communication, and sensors programs. He was the
founder of SysAir from 2002 to 2006, and developed first PC-based WCDMA baseband solution. From 2000 to
2002, He was the director of software development for Wiscom Technologies, a venture-capital funded start-up,
designing WCDMA baseband chip for cellular phone. From 1996 to 2000, He was a distinguished Member of
Technical Staff with Lucent Bell Labs and contributed to the NTT DoCoMo deliverables for commercial and
multi-millions R&D contracts. In early nineties, He worked at AT&T Bell Labs, and co-founded internet startup. He
received Ph.D. in Computer Science from the Ohio State University and undergraduate degree in Electrical
Engineering.
Thomas Mak, is currently a project manager of Technical Management Division of WIN-T, US Army, and serves as
a technical authority to provide directions to a group of senior researchers and system engineers from FFRDCs and
DoD contractors. He is the point of contact of US Army Terminal Program Office for TSAT/HC3/WIN-T system
integration. He represents Army/WIN-T for interfacing WIN-T routing, QoS, and tactical radios performance
requirements within GIG.
Prior to WIN-T, he worked for US Army CECOM and CERDEC Space and Terrestrial Communication Directorate.
He has over 20 years of program acquisition experience leading to contract awards for major military satellite and
commercial programs.
He has 10 years commercial broadband network deployment experience with AT&T and Lucent, served as a senior
project engineer for successfully deployed AirTel, Global Crossing, and Qwest networks. He received master
degrees of electrical engineering and mechanical engineering.
N2 − Technical Session: Network Solution and Performance Enhancement
Session Chair
Zhuangbo (Bo) Tang
11100 Johns Hopkins Road
Laurel MD 20723-6099
z.bo.tang@jhuapl.edu
BIAOGRAPHY
Zhuangbo Tang is currently a senior professional staff member at Applied Physics Laboratories of Johns
Hopkins University. Before that he worked at AT&T laboratories and Tellium (a start up company). His
academic experiences include a Post-doc position at Harvard University and a faculty position at Hong
Kong University of Science and Technology. He has been conducting research in the areas of
communication networks (IP, optical, wireless, Satcom), sensor networks, control systems, optimization,
and stochastic systems modeling/analysis/simulation.
N2 − Technical Session: Network Solution and Performance Enhancement
IP Fast Reroute for Shared Risk Link Group Failure
Recovery
Kang Xi
Polytechnic Institute of New York University
6 Metrotech Center, Brooklyn, NY 11201
kxi@poly.edu
ABSTRACT: Failure recovery in IP networks is critical to high-quality service provisioning. In IP over wavelength
division multiplexing (WDM) networks, a fiber carries multiple IP logical links. When a fiber fails, all the logical
links it carries are disconnected simultaneously. This is called a shared risk link group (SRLG) failure. Recovery
from SRLG failures using route recalculation could lead to long service disruption. In this paper, we present a
scheme called multi-section shortest path first (MSSPF) that achieves ultra fast recovery from SRLG failures.
MSSPF performs all the recovery related calculations in advance. On the detection of an SRLG failure, the affected
IP packets are detoured to their destinations through pre-calculated paths to avoid failed links. We prove that
MSSPF guarantees 100% recovery from SRLG failures and causes no permanent loops. In particular, the scheme
has low complexity and can be implemented in today's networks running link-state routing protocols, e.g., open
shortest path first (OSPF). The performance of our scheme is validated with a variety of practical and randomly
generated topologies.
BIAOGRAPHY: Dr. Kang Xi got his BSEE, MSEE and Ph.D. in Electronic Engineering from Tsinghua University
(Beijing, China) in 1998, 2000 and 2003, respectively. From 2004 to 2005 he worked as research associate as Osaka
University (Osaka, Japan). Since 2005 he has been faculty of Electrical and Computer Engineering at Polytechnic
University. His research interests including high speed networks, network resilience, traffic engineering, and
topology design.
N2 − Technical Session: Network Solution and Performance Enhancement
View-Upload Decoupling: A Redesign of Multi-Channel
P2P Video Systems
Yong Liu
Polytechnic Institute of NYU
5 MetroTech Center, Brookly, NY, 11201
mailto:yongliu@poly.edu
ABSTRACT: P2P video streaming is becoming an alternative IPTV solution with low server infrastructure cost. In
current multi-channel live P2P video systems, there are several fundamental performance problems including
exceedingly-large channel switching delays, long playback lags, and poor performance for less popular channels.
These performance problems primarily stem from two intrinsic characteristics of multi-channel P2P video systems:
channel churn and channel-resource imbalance. In this paper, we propose a radically different cross-channel P2P
streaming framework, called View-Upload Decoupling (VUD). VUD strictly decouples peer downloading from
uploading, bringing stability to multichannel systems and enabling cross-channel resource sharing. We propose a set
of peer assignment and bandwidth allocation algorithms to properly provision bandwidth among channels, and
introduce substream swarming to reduce the bandwidth overhead. We evaluate the performance of VUD via
extensive simulations as well with a PlanetLab implementation. Our simulation and PlanetLab results show that
VUD is resilient to channel churn, and achieves lower switching delay and better streaming quality. In particular, the
streaming quality of small channels is greatly improved.
BIAOGRAPHY: Dr. Yong Liu received his bachelor and master degree from the University of Science and
Technology of China, in 1994 and 1997 respectively. He graduated with Ph.D degree from ECE Dept. at University
of Massachusetts, Amherst in 2002. From February 2002 to February 2005, he worked as a Postdoc in Computer
Networks Research Group at UMass. In March 2005, he joined the ECE department of Polytechnic University as an
assistant professor. His current research interest includes: P2P systems, overlay networks, network measurement and
robust network design. More information about his research and teaching is available at:
http://eeweb.poly.edu/faculty/yongliu/
N2 − Technical Session: Network Solution and Performance Enhancement
Admission Control for VoIP Calls with Heterogeneous
Codecs
Xiaowen Mang, Yonatan Levy, Carolyn Johnson
and David Hoeflin
AT&T Labs Research
200 Laurel Ave. Middletown NJ 07748, USA
ABSTRACT: As VoIP services proliferate due to rapid advances in technology and market demands, service
providers are aggressively looking for better ways to offer VoIP services to customers including support of different
codec technologies via a single network access point. This service flexibility introduces major traffic engineering
challenges, especially when supporting a mix of data and voice services. In order to guarantee individual source’s
call blocking probability, intelligence must be added to the call admission process. To tackle the issue, we propose
an intelligent call blocking algorithm that is guided by a pre-defined blocking targets.
BIAOGRAPHY: Xiaowen Mang received her BS in Telecommunication Engineering from Beijing University of
Post and Telecommunications, Beijing, China; DEA (Diplome d'Etudes Approfondies) in Computer Methods for
Industrial Systems from University of Paris VI, Paris, France; and her Ph.D. in electrical and computer engineering
from Duke University, North Carolina, USA. She was awarded France Telecom Fellowship for her studies in France
and IBM Research Fellowship for her Ph.D. studies at Duke University. She is currently with AT&T Labs-Research.
During her tenure with AT&T she has worked on modeling and designing algorithms for a variety of AT&T
networks and systems. Dr. Mang holds eight US patents.
Yonatan Levy is executive director of the Network Design and Performance Analysis Division at AT&T Labs Research. Yoni has a vast experience in performance modeling and analysis and was responsible for efforts that led
to significant improvements in performance as well as to effective and reliable operation of network products and
services. Yoni holds several patents on dynamic network call distribution, packet network control and QoS, has
more than 20 publications, and in 2000 organized an ITC specialist seminar - the first international workshop
dedicated to IP traffic. Yoni has a Ph.D. in Mathematical Sciences from The Johns Hopkins University.
Carolyn Johnson is a Director of Quantitative Analysis at AT&T Labs Research in Middletown, NJ. Dr. Johnson
has extensive experience in telecommunications performance and reliability analysis. She has modeled voice services
performance, evaluated new network technologies, and designed congestion control algorithms. Her recent work
includes VoIP technologies and services, SIP overload control design, and IP network survivability analysis. Dr.
Johnson holds patents in overload controls, reliable element design, and priority mechanisms. Dr. Johnson joined
Bell Laboratories after receiving her Ph.D. in Mathematics from the University of Florida in 1980, and has been with
AT&T Labs since 1996.
David Hoeflin is a Technical Manager in the Network Design and Performance Analysis department of AT&T Labs
– Research. After receiving a Ph.D. in Mathematics (with a Minor in Statistics) from Iowa State University in
1984, David join AT&T Bell Labs to do performance and reliability analysis and remained in AT&T Labs to do
more of the same. More recently, Dave and his group have been doing reliability/performance in the area of IP
networks, services and products.
N2 − Technical Session: Network Solution and Performance Enhancement
Crossing a Non-Jackson Network (With or Without a Map)
Michael Tortorella
Rutgers University
Piscataway, NJ 08854
assurenet@verizon.net
ABSTRACT: This presentation considers the problem of computing cross-network values of performance
parameters whose values on single links are known in stochastic flow networks that have Markovian routing but
need not otherwise satisfy the Jackson network conditions. The method relies on generalizing the traffic equation
to path-additive functions of the flow in the network. A matrix-based approach provides computationally
convenient expressions for the results. These expressions involve the inverse of the matrix I  R where R is the
routing matrix of the network. Because routing in IP networks is address-based, not Markovian, we provide three
models for approximating address-based routing in a network model with Markovian routing.
BIOGRAPHY: Dr. Tortorella is a leading communications industry expert in reliability management, engineering,
modeling, and life data analysis. During a 26-year career at Bell Laboratories he was responsible for research and
implementations in fundamental system, network, and service reliability engineering methodologies as well as for
management of reliability in such critical projects as the SL-280 undersea cable system, the world's first application
of fiber-optic technology in an intercontinental, undersea system. He played a major role in many AT&T and
Lucent product reliability studies, culminating in the creation of CADRE, a reliability modeling system for circuit
packs that encompasses circuit simulation, thermal analysis, and uncertainty modeling in a single package that is
fully integrated with computer-aided design systems used for circuit pack creation.
Formerly technical manager and a Distinguished Member of Technical Staff in the Design for Reliability Processes
and Technologies Group and Next Generation Networks Reliability Group in Bell Laboratories, Dr. Tortorella is
now a research professor of industrial and systems engineering at Rutgers University. In addition to teaching
courses in industrial engineering and statistics, he maintains a robust research program that includes investigations
into how the stochastic flows in an IP network determine the performance and reliability of services carried on those
networks, developing modeling frameworks for control of IP networks under stressed conditions, and foundational
issues in queueing theory. Additional current research interests include stochastic flows, network performance,
management, and control, stochastic processes and their applications to reliability, life data analysis, and
next-generation networks, as well as design for reliability methods and technologies. Dr. Tortorella has published
extensively in these areas. At Bell Labs, his responsibilities included systems and reliability engineering for
next-generation networks (voice, wireless, and data). He received the Ph. D. degree in mathematics from Purdue
University in 1973. He is Advisory Editor for Quality Technology and Quantitative Management, where he has
worked to increase the number of publications pertaining to the communications industry. He was formerly Area
Editor for Reliability Modeling and Optimization for the IIE Transactions on Reliability and Quality Engineering
and was Guest Editor of a recent issue on Reliability Economics. He also served an Associate Editor of Naval
Research Logistics and was Guest Editor for a recent issue on Computations in Networks.
N3 − Technical Session: Network Reliability and Security
Session Chair
Mohcene Mezhoudi
600-700 Mountain Avenue
Murray Hill, NJ 07974
btang@alcatel-lucent.com
BIAOGRAPHY
Dr. M. Mezhoudi got his M.S.E.E. and Ph.D. From Stevens Institute of Technology, New Jersey. After
teaching at Stevens as an assistant professor and running the Optical Communications laboratory at
Stevens, he joined Bell Laboratories as a Member of Technical Staff in 1995. In 1999, he became a
Distinguished Member of Technical Staff. He then was promoted to Consultant Member of Technical Staff.
Dr. Mezhoudi has several publications in technical journals and conferences. He was awarded twice the
Bell Lab President award. His current research involves optical network transport, packet/optical network
switching and routing optimization techniques and reliability.
N3 − Technical Session: Network Reliability and Security
Simple Security Using Flow Data
Kenichi Futamura
AT&T, Inc.
200 S. Laurel Ave, Middletown, NJ 07748
futamura@att.com
ABSTRACT: Malware attacks have caused hundreds of billions of dollars in economic damage worldwide
yearly, and attackers are becoming smarter. We examine techniques for detecting security attacks in a
network using flow-level data. While worm exploits may be difficult to detect due to the wide range of
payloads, the propagation phase of a worm is generally much easier to recognize. We examine this step
and present one simple method for detecting network worms with no previously known signatures.
BIOGRAPHY: Kenichi Futamura received M.S. degrees in Mathematics (1994) and Statistics (1994) and a
Ph.D. in Operations Research (1996) at Stanford University. Since joining AT&T Labs, he has
investigated various areas including credit risk management, performance analysis, network grooming,
access optimization, and internet security. His recent security efforts include developing various intrusion
detection tools for the AT&T Internet Protect platform, including WARD, a worm detection tool. He has
various publications in technical journals and conferences as well as patents in network security and other
areas. Currently, he is a Principal Technical Staff Member, working on anomaly detection, intrusion
correlation, and capacity planning.
N3 − Technical Session: Network Reliability and Security
Detection of Spam Hosts and Spam Bots
Using Network Flow Traffic Modeling
Willa Ehrlich, Danielle Liu, David Hoeflin and Anestis Karasaridis
AT&T Labs, 200 Laurel Avenue, NJ 07981
dliu@att.com
ABSTRACT: In this paper, we present an approach for detecting email spammers and spam bots based on SMTP
network flow statistics. Our approach consists of establishing SMTP traffic models of legitimate vs. spammer SMTP
clients and then classifying an "unknown" SMTP client with respect to his/her current SMTP traffic distance from
these models.
We illustrate this approach based on a case study with SMTP flow data collected from our
backbone network. We demonstrate that a periodicity effect exists for SMTP traffic initiated by legitimate SMTP
clients and that we can adjust the traffic model parameter values for this periodicity using Exponentially Weighted
Moving Average (EWMA) smoothing. Given adjusted model parameter values, we demonstrate the accuracy of this
approach in classifying known blacklisted and whitelisted SMTP clients. Finally, we present an application of our
email spammer classification algorithm for detecting spammers that belong to botnets (also known as spam bots)
and the interactions utilized by these spam bots for command-and-control.
BIAOGRAPHY: Danielle Liu received her Ph.D. in Industrial Engineering under the guidance of Professor Marcel
Neuts at University of Arizona in 1993. She was a visiting professor at Department of Electrical Engineering at Case
Western Reserve University for one year before joining Bell labs in 1994. Danielle has worked on various projects
in AT&T including Internet traffic characterization, IP QoS, WiMAX and IP security. She is currently working on
email SPAM detection and network capacity planning.
Dr. Liu is the author of over 20 papers on in the fields of Queuing Analysis of Telecommunications Systems,
Internet Traffic Characterization and IP security. She also served as an editor for the journal of Queueing Systems:
Theory and Applications.
N3 − Technical Session: Network Reliability and Security
Network Vulnerability Identification via Network
Interdiction Research
Jose Emmanuel Ramirez-Marquez
Stevens Institute of Technology
School of Systems & Enterprises, Hoboken, NJ, 07030
jmarquez@stevens.edu
ABSTRACT: In many illegal or terrorist activities, networks are set up by the perpetrators to conduct their
operations, such as smuggling goods, contraband, and people across borders or ports, spreading CBRNE in an area,
etc. In order to potentially interdict these networks in a successful and cost effective way, multi‐objective
evolutionary algorithms are being developed to overcome issues of current techniques (e.g., the assumption that a
link interdiction will always disrupt network flow or a single criterion is used for optimization of a network). The
applications of this research go well beyond illegal activities and may be applied to any service system (i.e. electric
distribution systems, communication systems). This presentation will discuss how network interdiction can be used
to describe metrics such as vulnerability, resiliency and restoration response.
BIAOGRAPHY: Dr. Jose Emmanuel Ramirez-Marquez is an Assistant Professor of the School of Systems &
Enterprises at Stevens Institute of Technology. A former Fulbright Scholar, he holds degrees from Rutgers
University in Industrial Engineering (Ph.D. and M.Sc.) and Statistics (M.Sc.) and from Universidad Nacional
Autonoma de Mexico in Actuarial Science. His research efforts are currently focused on the reliability analysis and
optimization of complex systems, the development of mathematical models for sensor network operational
effectiveness and the development of evolutionary optimization algorithms. In these areas, Dr. Ramirez-Marquez
has conducted funded research for both private industry and government. Also, he has published more than 50
refereed manuscripts related to these areas in technical journals, book chapters, conference proceedings and industry
reports. Dr. Ramirez-Marquez has presented his research findings both nationally and internationally in conferences
such as INFORMS, IERC, ARSym and ESREL. He is an Associate Editor for the International Journal of
Performability Engineering and currently serves as director of the QCRE division board of the IIE and is a
member of the Technical Committee on System Reliability for ESRA.
N3 − Technical Session: Network Reliability and Security
End-to-End Service Reliability Considerations for
Converged Telecommunication Networks
Xuemei Zhang and Carolyn R. Johnson
AT&T Labs
200 St. Laurel Ave., Middletown, NJ 07748
xuemei.zhang@att.com; carolyn.johnson@att.com
ABSTRACT: Reliability metrics and modeling techniques have been successfully used to analyze the reliability and
availability of a system, which typically consists of some hardware platform and software running on top of the
hardware platform. In the telecommunications applications, system level reliability evaluation and improvement
activities are better-understood. However, measurements for assessing reliability of complex networks still need to
be better defined and understood. This becomes particularly essential for the mordent telecommunications networks
are getting more complicated with diversified technologies, multimedia services and evolving infrastructure. Besides
the traditional voice application, today’s telecommunication networks support multimedia applications that blended
text, voice and video services. These solutions deal with different technologies (e.g., wireless and wireline) and
involve network elements from different vendors. Moreover, different applications (e.g., games, gambling, bank
transactions, etc.) can have very different reliability requirements and characteristics. Techniques to analyze the
end-to-end service reliability of networks of such complexity and size are not well established. This paper discusses
reliability considerations in estimating and analyzing the end-to-end reliability of complicated telecommunications
network solutions.
BIAOGRAPHY:
XUEMEI ZHANG received her Ph.D. in Industrial Engineering and her Master of Science degree in Statistics from
Rutgers University, New Brunswick, New Jersey. Currently she is a principle member of technical staff in the
Network Design and Performance Analysis Department in AT&T Labs. Prior to joining AT&T Labs, she has worked
in the Performance Analysis Department and the Reliability Department in Bell Labs in Lucent Technologies (and
later Alcatel-Lucent), in Holmdel, New Jersey. She has been working on reliability and performance analysis of wire
line and wireless communications systems and networks. Her major work and research areas are system and
architectural reliability and performance, product and solution reliability and performance modeling, and software
reliability. She has published more than 30 journal and conference papers. She has 6 awarded and pending patent
applications in the areas of system redundancy design, software reliability, radio network redundancy, and end-to-end
solution key performance and reliability evaluation. Dr. Zhang is the recipient of a number of awards and scholarships,
including the Bell Labs President's Gold Awards in 2002 and 2004, Bell Labs President's Silver Award in 2005, and
Best Contribution Award 3G WCDMA in 2000 and 2001.
CAROLYN JOHNSON is a Director of Quantitative Analysis at AT&T Labs Research in Middletown, NJ. Dr.
Johnson has extensive experience in telecommunications performance and reliability analysis. She has modeled voice
services performance, evaluated new network technologies, and designed congestion control algorithms. Her recent
work includes VoIP technologies and services, SIP overload control design, and IP network survivability analysis. Dr.
Johnson holds patents in overload controls, reliable element design, and priority mechanisms. Dr. Johnson joined
Bell Laboratories after receiving her Ph.D. in Mathematics from the University of Florida in 1980, and has been with
AT&T Labs since 1996.
M1 − Technical Session: New Trends in Multimedia Technologies
Session Chair
Junlan Feng
AT&T Labs Research
Florham Park, New Jersey, USA
junlan@research.att.com
BIOGRAPHY
Dr. Junlan Feng is a principal research member at AT&T LABS RESEARCH. She received her Ph.D in Acoustics
from Chinese Academy of Sciences in 2001 and joined AT&T in the same year. Dr. Feng's research interest lies in
several technical areas including web mining, question answering, natural language understanding, machine
learning, information extraction, information retrieval, speech recognition, and spoken dialog management. Dr.
Feng holds dozens of issued or pending U.S. patents and has authored scores of publications.
M1 – New Trends in Multimedia Technologies
Natural Language Understanding on Mobile Voice Search
Junlan Feng
AT&T Labs Research
Florham Park, New Jersey, USA
junlan@research.att.com
ABSTRACT: Mobile voice-enabled search is emerging as one of the most popular applications abetted by the
exponential growth in the number of mobile devices. The automatic speech recognition (ASR) output of the voice
query is parsed into several fields and search is performed on a text or a database. In order to improve the
robustness of the query parser to noise in the ASR output, we extend our query parser powered by natural language
techniques to exploit multiple hypotheses from ASR, in the form of word confusion networks, in order to achieve
tighter coupling between ASR and query parsing. We observed improved accuracy of the query parser. We further
investigate the results of this improvement on search accuracy. Word confusion-network based query parsing
outperforms ASR 1-best based query-parsing by 2.7% absolute and the search performance improves by 1.8%
absolute on one of our data sets.
BIOGRAPHY: Dr. Junlan Feng is a principal research member at AT&T LABS RESEARCH. She received her
Ph.D in Acoustics from Chinese Academy of Sciences in 2001 and joined AT&T in the same year. Dr. Feng's
research interest lies in several technical areas including web mining, question answering, natural
language understanding, machine learning, information extraction, information retrieval, speech recognition, and
spoken dialog management. Dr. Feng holds dozens of issued or pending U.S. patents and has authored scores of
publications.
M1 – New Trends in Multimedia Technologies
Perceptual Quality Evaluation of Transmitted Videos
Tao Liu
Polytechnic Institute of New York University
LC 220, 5 MetroTech Ctr, Brooklyn, NY 11201
taoliu_bit@hotmail.com
ABSTRACT: Due to the rapid development of communication technologies nowadays, there is an increasing
demand for multimedia contents, such as videos, to be encoded with various codecs and transmitted over various
networks. Since human eyes are the very end users, the perceptual quality plays a key role in designing image/video
storage and transmission systems. Although subjective evaluation may be the only way to obtain the quality
assessment closest to the “true” value, it is extremely expensive to perform, and even not feasible in some
circumstances. Therefore, effective objective quality evaluation is of significant importance.
However, the evaluation of such videos is a highly challenging and complicated problem. Generally speaking, the
qualities of received videos are degraded at different levels, depending on both the choice of compression methods
and channel conditions. Additionally, the contents of transmitted videos, such as motion and saliency, also greatly
impact on the perceived quality too.
In order to solve this problem, we investigate each of the aforementioned quality-affecting elements individually by
designing and performing a series of subjective tests. And by taking advantage of several attributes of human visual
system, we finally propose our perceptual quality metrics which are shown that, from our subjective test data, they
can automatically and accurately predict the quality of the addressed videos.
BIOGRAPHY: Tao Liu is a Ph.D student in Electrical & Computer Engineering Department, Polytechnic Institute
of New York University, Brooklyn, NY, where he also received his Master degree in Electrical Engineering in 2007.
He earned his Bachelor degree in Electrical Engineering from Beijing Institute of Technology, Beijing, China, in
2004.
He joined in the Image Processing Lab at ECE department in Poly in 2004. In the summer of 2007, he worked as an
intern at Thomson Corporate Research, Princeton, NJ. From 2008, he participated in the collaborative work between
Poly and AT&T Labs- research, Florham Park, NJ. His current research interests include image analysis and
processing, pattern recognition, and perceptual video quality evaluation and enhancement.
M1 – New Trends in Multimedia Technologies
Multimedia Concept Detection: Cross Concept and Cross
Modality
Wei Jiang
Columbia University
1300 S.W.Mudd, 500 West 120th Street, New York, NY 10027
Wj2122@columbia.edu
ABSTRACT: Semantic indexing of images and videos becomes increasingly important. Traditional efforts classify
each concept individually, based on visual features such as color, texture, shape, etc.. Besides visual appearances,
other modalities, e.g., audio signals, provide useful information to help semantic classification, and how to utilize
information from multiple modalities is a very interesting issue. On the other hand, semantic concepts usually do not
occur in isolation and inter-conceptual relationship can help detect individual concepts. My talk focuses on effective
multimedia concept classification via cross-modality learning and cross-concept learning. The covered topics
include: joint semi-supervised learning of feature subspace and SVM classifier, which falls into the Early Fusion of
audio and visual feature representations by processing individual concepts separately; context-based concept fusion
that falls into the Late Fusion of audio-based and visual-based concept detectors with the setting of cross-concept
learning. Experiments over the Kodak’s consumer benchmark data set demonstrate significant performance
improvements..
BIOGRAPHY: Wei Jiang is a Ph.D student in the Electrical Engineering Department of Columbia University. She
graduated from Tsinghua University in China with a B.S. in June 2002 and a M.S. in June 2005, both in Department
of Automation. She had worked for IBM T.J. Watson and Eastman Kodak Company as a summer intern in 2008 and
2007, respectively. She also had been a visiting student in Microsoft Research Asia from 2003 to 2004. Her main
research interests are content-based image and video classification, indexing, and search, image analysis, and
machine learning. She is currently working with Professor Shih-Fu Chang studying semantic concept classification
with cross-modality learning, cross-concept learning, and cross-domain learning.
M1 – New Trends in Multimedia Technologies
Modeling Rate and Perceptual Quality of Scalable Video
and Its Application in Scalable Video Adaptation
Zhan Ma
Polytechnic Institute of New York University
LC 220, 5 MetroTech Ctr, Brooklyn, NY 11201
zhan.ma@gmail.com
ABSTRACT: Our work investigates the impact of frame rate and quantization on the bit rate and perceptual quality
of a scalable video with temporal and quality scalability. We propose a rate model and a quality model, both in
terms of the quantization stepsize and frame rate. Both models are developed based on the key observation from
experimental data that the relative reduction of either rate and quality when the frame rate decreases is quite
independent of the quantization stepsize. This observation enables us to express both rate and quality as the product
of separate functions of quantization stepsize and frame rate, respectively. The proposed rate and quality models are
analytically tractable, each requiring only two content-dependent parameters. Both models fit the measured data
very accurately, with high Pearson correlation. We further apply these models for rate-constrained bitstream
adaptation, where the problem is to determine the optimal combination of quality and temporal layers that provides
the highest perceptual quality for a given bandwidth constraint.
BIOGRAPHY: Zhan Ma was born in China on September 20, 1982. He received the B.S. and M.S. degrees in
Electrical Engineering from Huazhong University of Science and Technology, Wuhan, China, in 2004 and 2006
respectively. During the period of pursuing the M.S. degree, he had joined national digital audio and video
standardization
(AVS) workgroup to participate into standardizing the video coding standard in China. Since
September 2006, he has been a Ph.D. candidate at the Dept. of Electrical and Computer Engineering in Polytechnic
Institute of New York University, Brooklyn, NY, under the guidance of Professor Yao Wang. From May 2008 to
May 2009, he was an intern in Corporate Research, Thomson Inc., NJ. He mainly focused on the power, rate, and
perceptual quality modeling of the scalable video, and applications to the scalable video adaptation. He was the
recipient of the 2006 Special Contribution Award of the national digital audio and video standardization workgroup,
China for his contribution in standardizing the AVS Part 7 for mobile application.
M2 –Intelligent Multimedia Processing
Session Chair
Rong Duan
AT&T Labs Research
Florham Park, New Jersey, USA
rongduan@att.com
BIOGRAPHY
Rong Duan received her B.S and M.S in Computer Science and expects to receive her Ph.D. in Computer
Engineering in May, 2007. She joined AT&T Labs in 1998 and is currently a member of Applied Data Mining
Group. Her research interests include data mining with applications in image/video analysis, business intelligence,
and marketing. In particular, her research dissertation investigates supervised and semi-supervised learning methods
in pattern recognition problems. She is a member of INFORMS and IEEE and currently serves as the secretary and
treasurer of the INFORMS Data mining section.
M2 – Intelligent Multimedia Processing
Attack Estimation in Multimedia Contents Sharing
Systems
Wei Wang
Dept. of Electrical and Computer Engineering
Stevens Institute of Technology
Hoboken, NJ 07030
wwang3@stevens.edu
ABSTRACT: Nowadays, peer-to-peer (P2P) systems make multimedia sharing applications dominant the Internet
traffic. As a consequence, multimedia contents pollution became a serious security problem on the global Internet. It
is generally impractical to exhaustedly search for the pollution sources due to the huge amount of users in the P2P
network. At the same time, because the infrastructure of a P2P network is flat, there are no centered servers that
could employ the access control over contents published or shared by every live user. In this paper, we first review
most pervasive attacks in current P2P systems, such as poisoning and content spam pollution. Then we propose a
passive scheme by deploying a very few agent nodes to detect malicious content alternation users. Simulation results
show that after applying our detection system, we can roughly estimate the proportion of malicious users who apply
content alternation attacks in a P2P network.
BIOGRAPHY: Wei Wang got her B.S. and M.S. from Huazhong University of Science and Technology in 2001 and
2004 separately. From 2006 to now, she is a Ph. D. student in ECE department of Stevens Institute of Technology.
Her research interests mainly concentrate on network security, social networks and overlay networks.
M2 – Intelligent Multimedia Processing
Adaptive Mean Shift for Target Tracking in FLIR Imagery
Yafeng Yin
ECE Department
Stevens Institute of Technology
Castle Point on Hudson, Hoboken, NJ 07030
yyin1@stevens.edu
ABSTRACT: Reliable tracking of targets in the Forward-Looking Infrared (FLIR) imagery is a challenging work in
the computer vision, since IR images usually have extremely low contrast and inconspicuous difference between
targets and background. In this paper, we present a novel adaptive mean shift tracker for tracking moving targets in
the FLIR imagery, captured from an airborne moving platform. First, each target’s initial position is manually
marked to initialize the adaptive mean-shift based tracker. For each target, multiple different features are extracted
from both the targets and background during tracking, and an on-line feature ranking method is deployed to
adaptively select the most discriminative feature for the mean-shift iteration. In addition, to compensate the motion
of the moving platform, a block matching method is applied to compute the motion vector, which will be used in the
RANSAC algorithm to estimate the affine model for global motion. We test our method on the AMCOM FLIR data
set, the result indicate that our Adaptive mean-shift tracker can track each target accurately and robustly.
BIOGRAPHY: Yafeng Yin received his bachelor degree in the department of Automatic Control from Beijing
Institute of Technology in 2005. He is Currently a PhD candidate at the ECE department of Stevens Institute of
Technology. His research interests mainly involve image analysis and machine learning, with application on
video-base people tracking and human behavior analysis.
M2 – Intelligent Multimedia Processing
Fast Rerouting for IP Multicast in Managed IPTV
Networks
Dongmei Wang
AT&T Labs Research
Florham Park, New Jersey, USA
mei@research.att.com
ABSTRACT: Recent deployment of IP based multimedia distribution, especially broadcast TV distribution has
increased the importance of simple and fast restoration during IP network failures for service providers. The
restoration mechanisms currently adopted in IP networks use either IGP re-convergence (which could be too slow
for multimedia content distribution) or IP/MPLS fast reroute. Both would increase router configuration and network
operation complexity as well as human-errors. Also the service provider IP/MPLS networks are mainly tuned to
support unicast traffic, and some of the multicast functions are not fully supported yet. In this paper, we propose and
evaluate a simple but efficient method for fast rerouting of IP multicast traffic during link failures in managed IPTV
networks. More specifically, we devise an algorithm for tuning IP link weights so that the multicast routing path and
the unicast routing path between any two routers are failure disjoint, allowing us to use unicast IP encapsulation for
undelivered multicast packets during link failures. We demonstrate that, our method can be realized with minor
modification to the current multicast routing protocol (PIM-SM). We run our prototype implementation in Emulab
which shows our method yields to good performance.
BIOGRAPHY: Dr. Dongmei Wang received her M.S from Beijing Normal University in 1995, and PhD from the
college of William and Mary in 2000. Since then, she joined AT&T research lab and has been working on network
related research topics, from optical layer to application, architectures to protocols, algorithms to simulation,
provisioning to restoration. Most recently, she has been focusing on IPTV related problems. She has been the author
of more than 30 research papers and filed 15 patents.
M2 – Intelligent Multimedia Processing
Learning Visual Features via the Neighbor-Constrained
Hierarchical Network
Yuhua Zheng
Embedded Systems and Robotics Laboratory
ECE, Stevens Institute of Technology, Hoboken, NJ, 07030
yzheng1@stevens.edu
ABSTRACT: Learning and recognition of visual objects has been a central research topic in computer vision for
several decades. Many different models and approaches have been proposed to represent and learn the visual
features, among which the hierarchical network, like convolutional neural network, HMAX model and deep
Boltzmann machine, has demonstrated the ability of representing the multiple-level patterns of objects. In this paper,
a multi-layer neighbor-constrained network model (NCHN) is proposed to represent hierarchical features for visual
object representation. The connections in this network are constrained by the neighborhoods of nodes, which reflect
the topologies and dependencies of different parts of the object. Compared with the fully-connected network, the
number of connections is reduced and the spatial relationships are kept. By applying a learning algorithm of
minimizing contrastive divergence, this model is able to learn complex feature structures from unlabelled data. More
specifically, this model can provide hierarchical feature structures of the object of interest. The lower layer
expresses more detailed appearance features while the feature represented by the higher layer is more compact and
abstract. The experimental results demonstrate the efficiency of the learning capability of the proposed model and
the feature hierarchies from the model for reconstruction.
BIOGRAPHY: Yuhua Zheng is now a PhD student of the department of Electronic and Computer Engineering,
Stevens Institute of Technology, Hoboken, NJ. He got both bachelor and master degree of electronic engineering
from Huazhong University of Scienc and Technology in 2001 and 2004 respectively. Then since 2004, he worked
for Alcatel Shanghai-Bell as an engineer on multimedia broadcast projects.
Yuhua Zheng’s research interests include computer vision, pattern recognition and machine learning, especially on
visual object recognition and tracking, complex pattern representation and clustering, and hierarchical network
evolving with bio-inspired algorithms.
M3 – Novel Multimedia Applications
Session Chair
Xiang Zhou
Siemens
51 Valley Stream Parkway, Malvern, PA 19355
Xiang.zhou@siemens.com
BIOGRAPHY
Xiang "Sean" Zhou conducted his PhD study (1998-2002) at Beckman Institute for Advanced Science and
Technology at University of Illinois at Urbana Champaign (UIUC). He worked as a researcher (2002) and later
project manager (2004) at Siemens Corporate Research in Princeton, New Jersey. Since June 2005, he has been
working for Siemens Medical Solutions, as a senior staff scientist, a program manager, and now a senior manager at
the Computer Aided Diagnosis and Knowledge Solutions Group in Malvern, Pennsylvania.
He publishes in IEEE Transactions on Medical Imaging, IEEE Transactions on Pattern Analysis and Machine
Intelligence, IEEE Multimedia, IEEE Transactions on Circuits and Systems for Video Technology, Optical
Engineering, ACM Multimedia Systems Journal, etc., and leading international conferences. He is the principle
author of the book "Exploration of Visual Data", published by Kluwer Academic Publishers in 2003. He was the
recipient of several top scholarships and awards from Tsinghua University. While studying in UIUC, he was
awarded the 2001 M. E. Van Valkenburg Fellowship, an award given to one or two PhD students in the ECE
department of UIUC each year "for demonstrated excellence in research in the areas of circuits, systems, or
computers."
M3 – Novel Multimedia Applications
Player Highlighting and Team Classification in Broadcast
Soccer Videos for the Next Generation TV
Yu Huang
Multimedia Content Networking, Huawei Technologies (USA)
400 Somerset Corporate Blvd, Suite 602, Bridgewater, NJ 08807
e-mail: yhuang@huawei.com
ABSTRACT: The coming next generation TV is making the viewing experience more interactive and personalized,
for example, the popular IPTV. In this paper, we discuss a scenario about a rich media interactive TV application for
IPTV, mainly interaction with objects of interests in the sports programs. We propose a framework for player
segmentation, tracking, highlighting and team classification in soccer game videos. In player segmentation, playfield
modeling is realized by a semi-supervised method, combining the Gaussian distribution model with the dominant
color detection. In player tracking, a modified mean shift-based method is proposed, which takes into account the
soft constraints from the foreground map, to handle the fast moving players. In the tracking process, scale change
and drifting artifacts are two of critical issues. In our tracking module, a discriminant similarity metric de-weighted
by the surrounding background distribution is applied to handle the “shrinkage” problem in scale adaptation;
meanwhile, a conservative way to handle the object’s appearance variation is proposed, which updates the target
model by aligning with its initial. In our experiment demonstration, two use cases are presented: one is player
highlighting based on segmentation and tracking; and the other is team classification with a bi-histogram matching
scheme.
BIAOGRAPHY: Dr. Huang got his B.S. from Xi’an Jiao Tong University, M.S. from Xidian University (formerly
Xi’an Institute of Telecommunications and Engineering) and Ph. D. from Beijing Jiao Tong University. After being
the researcher and lecturer at Tsinghua University for two years, he was awarded the Alexander von Humboldt
Research Fellowship in 1999, hosted at Institute of Pattern Recognition, University of Erlangen-Nuremberg
(Bayern, Germany). During 2000-2003 he was a Postdoctoral Research Associate of Prof. Thomas S. Huang,
Beckman Institute of Advanced Science & Technology, University of Illinois at Urbana-Champaign. After working
as a R&D algorithm engineer at Rapiscan Systems Inc. (a subsidiary of OSI systems Inc.) for more than two years,
he became a Senior Member of Technical Staff at Corporate Research of Thomson Multimedia Inc. in 2005-2008.
Since April 2008, he has been a Senior Researcher of Multimedia Content Networking at Core Network Research
Dept., Huawei Technologies (USA).
Dr. Huang has published more than 30 academic papers in international conferences and journals, and he has filed 8
US patents. His experience consists of signal/image processing, video analysis and mining, machine learning,
image-based rendering, data visualization, computer vision and human-computer interaction etc.
Dr. Huang is member of IEEE and ACM.
M3 – Novel Multimedia Applications
Computer Graphics Classification Based on Markov Process
Model
Patchara Sutthiwan*, Xiao Cai*, Yun Q. Shi*, Hong Zhang+
*
Dept. of Electrical and Computer Engineering, New Jersey Institute of Technology,
Newark, NJ
{ps249, xc27, shi}@njit.edu
+
Dept. of Computer Science, Armstrong Atlantic State University, Savannah, GA
hong@drake.armstrong.edu
ABSTRACT: In this work, a novel technique is proposed to identify computer graphics, employing second-order
statistics to capture the significant statistical difference between computer graphics and photographic images. Due to
the wide availability of JPEG images, a JPEG 2-D array formed from the magnitudes of quantized block DCT
coefficients is deemed a feasible input, but a difference JPEG 2-D array tells a better story about image statistics
with less influence from image content. Characterized by transition probability matrix (TPM), Markov process,
widely used in digital image processing, is applied to model the difference JPEG 2-D arrays along horizontal and
vertical directions. We resort to a thresholding technique to reduce the dimensionality of feature vectors formed
from TPM. YCbCr color system is selected because of its demonstrated better performance in computer graphics
classification than RGB color system. Furthermore, only Y and Cb components are utilized for feature generation
because of the high correlation found in the features derived from Cb and Cr components. The effectiveness of the
image feature vector is then evaluated by the classifier in the machine learning (ML) framework, such as Support
Vector Machines (SVM) classifier. Experimental works have shown that the proposed method outperforms the prior
arts by a distinct margin.
BIOGRAPHY:
Patchara Sutthiwan is a Ph.D. student of the Electrical and Computer Engineering Department at New Jersey
Institute of Technology. His research interests include visual signal processing, multimedia security and digital
image forensics. He received the B.Eng degree from Chulalongkorn University, Bangkok, Thailand and M.S. degree
from New Jersey Institute of Technology, both in Electrical Engineering, in 2001 and 2006, respectively.
Xiao Cai earned his Bachelor of Science in Information Engineering from Tianjin University, Tianjin, P.R. China in
2003 and Master of Science in Electrical Engineering from New Jersey Institute of Technology in January 2009. He
is currently an intern at Vitro Imaging Systems department at Abbott Point of Care Inc, Princeton. His research is
about image processing and machine learning.
Dr. Yun Qing Shi has joined the Department of Electrical and Computer Engineering at the New Jersey Institute of
Technology since 1987, and is currently a professor there. He obtained his B.S.degree and M.S.degree from the
Shanghai Jiao Tong University, Shanghai, China; his M.S. and Ph.D. degrees from the University of Pittsburgh. His
research interests include visual signal processing and communications, multimedia data hiding and security,
applications of digital image processing, computer vision and pattern recognition to industrial automation and
biomedical engineering, theory of multidimensional systems and signal processing.
Dr. Hong Zhang is currently a professor of Computer Science at Armstrong Atlantic State University. He received
his MS.EE in Electrical Engineering and Ph.D. in Mathematics from University of Pittsburgh. Dr. Zhang has
extensive experiences in both academia and industries. His research interests include graph theory, algorithms,
control theory, machine learning, computer graphics, image processing and biomedical applications.
M3 – Novel Multimedia Applications
An Adaptive Bottom Up Clustering Approach for Web
News Extraction
Jinlin Chen
Queens College, City Univ. of New York
65-30 Kissena Blvd., Flushing, NY, 11367
jchen@cs.qc.cuny.edu
ABSTRACT: An adaptive bottom up Web news extraction approach based on human perception is presented in this
paper. The approach simulates how a human perceives and identifies Web news information by using an adaptive
bottom up clustering strategy to detect possible news areas. It first detects news areas based on content function,
space continuity, and formatting continuity of news information. It further identifies detailed news content based on
the position, format, and semantic of detected news areas. Experiment results show that our approach achieves much
better performance (in average more than 99% in terms of F1 Value) compared to previous approaches such as Tree
Edit Distance and Visual Wrapper based approaches. Furthermore, our approach does not assume the existence of
Web templates in the tested Web pages as required by Tree Edit Distance based approach, nor does it need training
sets as required in Visual Wrapper based approach. The success of our approach demonstrates the strength of the
perception based Web information extraction methodology and represents a promising approach for automatic
information extraction from sources with presentation design for humans.
BIOGRAPHY: Dr. Chen got his B.S. and Ph.D. from Tsinghua University. After working at Microsoft Research
Asia for two years from 1999 to 2001, he joined Univ. of Pittsburgh as a Visiting Professor in 2002. In 2003, he
joined Queens College, City Univ. of New York and has been a faculty member at Computer Science Department
since then.
Dr. Chen’s research interests include data mining, information retrieval/extraction, Web information modeling and
processing. He received a highlight paper award in WWW2001 and best paper award in IEEE International
Conference on Digital Information Management 2006. He also holds four US patents.
M3 – Novel Multimedia Applications
Computer Aided Detection of Anatomical Primitives in
Medical Images and Its Applications
Xiang Zhou
Siemens
51 Valley Stream Parkway, Malvern, PA 19355
Xiang.zhou@siemens.com
ABSTRACT: Medical image retrieval applications pose unique challenges but at the same time offer many new
opportunities. On one hand, while one can easily understand news or sports videos, a medical image is often
completely incomprehensible to untrained eyes. On the other hand, semantics in the medical domain is much better
defined and there is a vast accumulation of formal knowledge representations that could be exploited to support
semantic search for any specialty areas in medicine.
In this talk, however, we will not dwell on any one particular specialty area, but rather address the question of how
to support scalable semantic search across the whole of medical imaging field: what are the advantages to take and
gaps to fill, what are the key enabling technologies, and the critical success factor from an industrial point of view.
In terms of enabling technologies, we discuss three aspects: 1. scalable image analysis and anatomical tagging
algorithms; 2. anatomical, disease, and contextual semantics, and their representations using ontologies; and 3.
ontological reasoning and its role in guiding and improving image analysis and retrieval.
More specifically, for scalable image analysis we present a learning- based anatomy detection and segmentation
framework using distribution-free priors. It is easily adaptable to different anatomies and different imaging
modalities. Examples of intelligent algorithms for medical imaging equipments (such as CT, MRI, and Ultrasound
machines) will be presented. For ontological representation of medical imaging semantics, we discuss the potential
use of FMA, RadLex, ICD, and AIM.
BIOGRAPHY: Xiang "Sean" Zhou conducted his PhD study (1998-2002) at Beckman Institute for Advanced
Science and Technology at University of Illinois at Urbana Champaign (UIUC). He worked as a researcher (2002)
and later project manager (2004) at Siemens Corporate Research in Princeton, New Jersey. Since June 2005, he has
been working for Siemens Medical Solutions, as a senior staff scientist, a program manager, and now a senior
manager at the Computer Aided Diagnosis and Knowledge Solutions Group in Malvern, Pennsylvania.
He publishes in IEEE Transactions on Medical Imaging, IEEE Transactions on Pattern Analysis and Machine
Intelligence, IEEE Multimedia, IEEE Transactions on Circuits and Systems for Video Technology, Optical
Engineering, ACM Multimedia Systems Journal, etc., and leading international conferences. He is the principle
author of the book "Exploration of Visual Data", published by Kluwer Academic Publishers in 2003. He was the
recipient of several top scholarships and awards from Tsinghua University. While studying in UIUC, he was
awarded the 2001 M. E. Van Valkenburg Fellowship, an award given to one or two PhD students in the ECE
department of UIUC each year "for demonstrated excellence in research in the areas of circuits, systems, or
computers."
M4 – Machine Learning for Information Management & Analysis
Session Chair
Yanjun Qi
Machine Learning Department, NEC Lab America, Inc
4 Independence Way, Suite 200, Princeton, NJ 08536
yanjun@nec-labs.com
BIOGRAPHY: Dr. Qi obtained her Ph.D. degree from School of Computer Science at Carnegie Mellon University
in 2008. She received her Bachelor with high honors (and also M.E. in the accelerated program) from Computer
Science Department at Tsinghua University in 2001. Currently, Dr. Qi is a postdoctoral scientist in the Machine
Learning Department at NEC Lab America, Princeton, NJ. Dr. Qi specializes in machine learning applications for
biological networks, video analysis and text mining. She is a member of ACM and IEEE.
M4 – Machine Learning for Information Management & Analysis
Metric-based Automatic Taxonomy Induction
Hui Yang
Language Technologies Institute
School of Computer Science
Carnegie Mellon University
5000 Forbes Ave, LTI, CMU, Pittsburgh, PA, 15213
huiyang@cs.cmu.edu
ABSTRACT: This talk presents a novel metric-based framework for the task of automatic taxonomy induction. The
framework incrementally clusters terms based on ontology metric, a score indicating semantic distance; and
transforms the task into a multi-criteria optimization based on minimization of taxonomy structures and modeling of
term abstractness. It combines the strengths of both lexico-syntactic patterns and clustering through incorporating
heterogeneous features. The flexible design of the framework allows a further study on which features are the best
for the task under various conditions. The experiments not only show that our system achieves higher F1-measure
than other state-of-the-art systems, but also re-veal the interaction between features and various types of relations, as
well as the interaction between features and term abstractness.
BIOGRAPHY: Hui Yang is a Ph.D. candidate in the Language Technologies Institute, School of Computer Science,
Carnegie Mellon University. She received her Master of Computer Science from School of Computer Science,
Carnegie Mellon University; and Bachelor of Computer Science from School of Computing, National University of
Singapore.
Hui Yang was one of the top 200 students in China in 1996, and hence received the Singapore Ministry of Education
Scholarship for her undergraduate study in National University of Singapore. After the undergraduate study, she was
offered the position as a junior instructor by National University of Singapore. During her academic career in
Singapore, she taught Artificial Intelligence, Multimedia Processing, and Software Engineering, as well as actively
conducted research in Question Answering and Multimedia Information Retrieval. Her work on Question
Answering participated in TREC 2002 and TREC 2003, and was the 2 nd best system for both years among systems
from all over the world.
Hui Yang’s research interests focus on text mining, information retrieval and machine learning. Her current research
includes automatic/semi-automatic ontology generation, human-guided machine learning, text analysis and
organization. Her earlier work includes near duplicate detection in large text corpora and the Web, question
answering, multimedia information retrieval, and opinion and sentiment detection. She has published more than 20
research papers in various conferences, including the top conferences, such as SIGIR, WWW, ACM Multimedia and
CIKM.
Hui Yang actively conducts professional service in her research field. She is the chair of student research paper for
the Digital Government (DG.O) conference in 2009. She is also the PC member for SIGIR 2008, DG.O 2009, and
reviewers for SIGIR 2004, SIGIR 2008. She organizes the Information Retrieval Seminar in Carnegie Mellon
University since 2006.
M4 – Machine Learning for Information Management & Analysis
Non-rigid Face Tracking with Enforced Convexity and
Local Appearance Consistency Constraint
Yang Wang
Siemens Corporate Research
755 College Road East, Princeton, NJ 08540
wangy@cs.cmu.edu
ABSTRACT: Accurate and consistent tracking of non-rigid object motion is essential in many computer graphics
and multimedia applications, especially dynamic facial expression analysis, such as facial expression recognition,
classification, detection of emotional states, etc. In this paper we present a new discriminative approach, based on
the constrained local model (CLM), to achieve consistent and efficient tracking of non-rigid object motion, such as
facial expressions. By utilizing both spatial and temporal appearance coherence at the patch level, the proposed
approach can reduce ambiguity and increase accuracy. More importantly, we show that the global warp update can
be optimized jointly in an efficient manner using convex quadratic fitting. Finally, we demonstrate that our approach
receives improved performance for the task of non-rigid facial motion tracking on the videos of clinical patients.
BIOGRAPHY: Dr. Wang received the Bachelor and Master degrees from Tsinghua University in 1998 and 2000,
respectively. He spent 2 years as a Postdoctoral Fellow in the Robotics Institute at Carnegie Mellon University from
2006 to 2008, after he obtained the Ph.D. degree from the Department of Computer Science at Stony Brook
University. Currently, Dr. Wang is a research scientist in the Integrated Data Systems Department at Siemens
Corporate Research, Princeton, NJ.
Dr. Wang specializes in non-rigid motion tracking in 2D videos and 3D medical images, facial expression analysis
and synthesis, and illumination modeling. He has published more than 20 papers in the top journals and conferences
in computer vision and graphics. He is a member of ACM and IEEE.
M4 – Machine Learning for Information Management & Analysis
Supervised Semantic Indexing
Bing Bai
NEC labs America, Inc.
4 Independence Way, Princeton, NJ 08540
bbai@nec-labs.com
ABSTRACT: We present a class of models that are discriminatively trained to directly map from the word content
in a query-document or document-document pair to a ranking score. Like Latent Semantic Indexing (LSI),
our models take account of correlations between words (synonymy, polysemy). However, unlike LSI our models are
trained with a supervised signal directly on the task of interest, which we argue is the reason for our superior results.
As the query and target texts are modeled separately, our approach is easily generalized to other retrieval tasks
such as cross-language retrieval as well. We provide an empirical study on retrieval tasks based on Wikipedia
documents, where we obtain state-of-the-art performance using our method.
BIOGRAPHY: Dr. Bai got his M.S. and Ph.D. from Rutgers University. He is currently a postdoctoral scientist in
the machine learning department of NEC labs America, INC.
M4 – Machine Learning for Information Management & Analysis
Probabilistic Knowledge Model for Document Retrieval
Shuguang Wang
University of Pittsburgh
210 S. Bouquet St., Sennott Square 5406, Pittsburgh, PA 15260
swang@cs.pitt.edu
ABSTRACT: The objective of our research is to find ways of enhancing information retrieval methods with the help
of domain knowledge, which can come from different sources. The basis of our knowledge model is a network of
associations among concepts defining various domain entities. This network can be extracted from the literature
corpus. The probabilistic model used to support inferences is built from this knowledge network with the help of the
link analysis methods. We propose to build a probabilistic knowledge model of relations among domain concepts
(defining entities) from literature corpus and exploit the model to improve the document retrieval. We test our
approach on biomedical documents retrieval problem and show that the new approach outperforms Lucene.
BIOGRAPHY: Mr. Wang graduated with B.Sc. and M.Sc. from National University of Singapore. He is currently
pursuing PhD in Intelligent Systems Program University of Pittsburgh. Before joining University of Pittsburgh, he
was a software developer in a Tropical Marine Science Institute, Singapore and BBN Technologies Boston.
Mr. Wang has received multiple scholarships from Ministry of Education Singapore during his undergraduate study,
and he also received fellowship from University of Pittsburgh.
Mr. Wang has participated in several projects in Question Answering, Machine Translation, Speech Recognition,
Database Indexing and Machine Learning. He currently studies the problem of learning probabilistic knowledge
model using link analysis from research literature and knowledge bases. This work has been applied into document
retrieval task and shows positive results.
Mr. Wang has published several papers in technical journals and conferences on database indexing, question
answering and document retrieval. He also served as a PC member in DG.O 2009 student research track.
Download