|
Alistair
Rendell |
Wanlei Zhou Melbourne, Australia Title: Effective Parallel Spam Filtering using Multiple Classifiers |
Stephan Olariu |
Alistair Rendell
Dr Rendell has a BSc
from Durham University and a PhD from the University of Sydney. He is widely
known for his work in the development, application and tuning of computational
science applications (particularly chemistry) to high performance scientific environments.
In late eighties he developed a highly vectorized and parallelized
coupled-cluster algorithms that achieved over 90% of the peak performance on an
8 processor Cray Y-MP. During the early 90.s he developed the first massively
parallel coupled cluster code, a code that now forms part of NWChem
computational chemistry program. This work pioneered the use of one-sided
communications for implementing pseudo shared memory on distributed memory
computers, concepts from this work are now included in the widely used Global
Arrays and Distributed Data Interface libraries.
From
1995-2000 he worked at the Australian National University (ANU) Supercomputer
Facility where he led a major collaborative project concerned with the
development, porting, and optimization of computational science software for
Fujitsu supercomputers. While this work was commercial in confidence, the
results from this work were incorporated into a variety of well know
computational packages (e.g. Gaussian, Amber, GAMESS etc).
In
2001 Dr. Rendell joined the faculty in the Computer Science Department. Since
then he has been promoting research and teaching in all aspects of high
performance computing. Within the Department he currently leads the Computer
Systems group comprising 15-20 members. He has active collaborations with a
variety of industry partners and is author of over 75 refereed publications.
For more information, refer to his
home page.
Title: Computational Science Applications for Modern Parallel
Architectures: Experiences, Challenges and Future Opportunities
Abstract: For many years we have been
developing computational science applications, particularly chemistry, for high
performance parallel systems. This
talk will highlight some of our recent work related to non-uniform memory
access (NUMA) parallel systems, the development and use of cluster-enabled
openMP environments, and the exploitation of novel parallel architectures such
as the Compute Unified Device Architecture (CUDA) available on NVIDIA GPUs and
the IBM Cell Broadband Engine (Cell BE).
On NUMA systems our goal is to build models that can be used to predict
performance as a function of memory and thread placement and that can be
exploited during runtime to enhance performance. For cluster OpenMP we have both
developed our own in-house software distributed memory system, but are also
working with Intel on their Cluster OpenMP (ClOMP) product. Results using ClOMP
for the Gaussian quantum chemistry code will be presented and discussed. For
the GPU and the Cell BE system we have been exploring the development of
molecular dynamics and machine learning applications on these two systems. In
this talk we will attempt to compare the two platforms, their relative merits
and difficulties.
Wanlei Zhou
Professor Wanlei Zhou
received his PhD degree from The Australian National University, Canberra, Australia,
in October 1991. He also received the DSc degree from Deakin University,
Victoria, Australia in 2002. He is currently the Chair Professor of Information
Technology and the Associate Dean (International), Faculty of Science and
Technology, Deakin University, Melbourne, Australia. His research interests
include distributed and parallel systems, network security, mobile computing,
bioinformatics and e-learning. Professor Zhou has published more than 170
papers in refereed international journals and refereed international
conferences proceedings. Since 1997 Professor Zhou has been involved
in
more than 50 international conferences as General Chair, Steering Chair, PC
Chair, Session Chair, Publication Chair, and PC member. Professor Zhou is a
member of the IEEE.
For more information, refer to his
home page.
Title: Effective Parallel Spam Filtering using Multiple
Classifiers
Abstract: Emails have now become an integral
part of everyday life and a prime means of communication tool for idea and
information exchange. However, along with the rapid growth of the Internet and
email, there has been a dramatic growth in spam in recent years. Spam is
commonly defined as unsolicited email messages and protecting email from
infiltration of spam is an important research and practical issue nowadays.
Spam filtering using classification algorithms has been successfully used in
practice, but with certain amount of false positive tradeoffs. False positive
is unacceptable in many cases as the loss of an important email could have
significant implications. This problem is mainly caused by the dynamic nature
of spam contents, sending strategies as well as diversification of the
classification algorithms. This
talk will introduce the innovative solutions carried out in my research group
for effective parallel spam filtering using multiple classifiers which will be
able to achieve high accuracy and overcome the false positive problem. The talk is divided into the following
parts:
1.
Current
spam filtering methods
2.
The
multi-classifier classification model
3.
Grey
list analysis and dynamic feature selection
4.
A
multi-core framework for multi-classifier email classification
Stephan Olariu
Professor Olariu has
held many different roles and responsibilities as a member of numerous
organizations and teams. Much of his experience has been with the design and
implementation of robust protocols for wireless networks and in particular
sensor networks and their applications. He is applying mathematical modeling
and analytical frameworks to the resolution of problems ranging from securing
communications, to predicting the behavior of complex systems, to evaluating
performance of wireless networks. His research interests are in the area of
complex systems enabled by large-scale deployments of sensors and more
specifically in securing systems of systems. Professor Olariu is a
world-renowned technologist in the areas of wireless networks, mobile
multimedia systems, parallel and distributed systems, parallel and distributed
architectures and networks. He was invited and visited more than 120
universities and research institutes around the world lecturing on topics
ranging from wireless networks and mobile computing, to biology-inspired
algorithms and applications, to telemedicine, to wireless location systems, and
security. He is the Director of the Sensor Networks Research Group at Old
Dominion University.
For more information, refer to his
home page.
Title:
An architecture for Traffic Incident
Detection
Abstract: Road and traffic safety can be
improved if the drivers have the ability to see further down the road and can
be informed of relevant traffic events, including collisions and
slow-downs. The recently proposed
VANETs (Vehicular Ad hoc Networks) are expected to enable both
vehicle-to-vehicle (V2V) and vehicle-to-roadside (V2R) communications.
Virtually all the papers published in the literature assume that V2V
communications will rely on a strong roadside infrastructure. Unfortunately,
the roadside infrastructure, is very likely to be the target of theft,
vandalism and other similar activities that will jeopardize their intended
functionality. Worse yet, one can easily contemplate a scenario where the
roadside infrastructure may be hacked and injected with malicious code,
rendering it not only useless but, downright dangerous.
However, all the VANET
systems proposed thus far are afflicted with serious security and privacy
problems. Indeed, the way current systems are set up, the driver of a car that
participates in the traffic will not be able to preserve their privacy and may
be subject to impersonation or Sybil attacks. The problem stems from the fact
that V2V communication can be traced back to an individual car. Even if several
pseudonyms are used, detecting the true identity of the driver and, therefore,
invading their privacy appears to be unavoidable.
In a sharp departure from
the common wisdom we propose to look at vehicle-to-vehicle (V2V) and
vehicle-to-roadside (V2R) communications from a different perspective. Instead
of relying on the roadside infrastructure that is vulnerable to attacks, we
propose to embed in the asphalt covering the surface of the roads sensor belts.
Each belt consists of a collection of pressure sensors, optionally equipped
with piezo-electric elements. The belts can detect and interact with passing
cars.
In this talk we discuss
in detail NOTICE our architecture for traffic incident detection and show that
it can be easily extended to cover many problems of interest in infotainment
and peer-to-peer content delivery. One important application of NOTICE is with
planned evaluations when optimal use must be made of available transportation
resources.