Date:
3/06/09
Title: Generalized
Integrated Learning Architecture
Abstract: In this talk we
present an ensemble architecture for learning complex problem solving
techniques from expert demonstration.
The key feature of our ``Generalized Integrated Learning Architecture"
(GILA) is a set of integrated learning and reasoning (ILR) systems
coordinated by a central meta-reasoning executive (MRE). Each ILR
learns independently from the same training example and contributes
to problem-solving in concert with other ILRs as directed by the MRE.
We describe the application of this novel learning and problem solving
architecture to the domain of airspace management, where multiple
requests for the use of airspace need to be reconciled and managed
automatically. Formal evaluations show that our system performs
as well as or better than humans after learning from the same training
data. Further, GILA outperforms any individual ILR run in isolation,
thus demonstrating the power of the ensemble architecture for learning
and problem solving.
Date: 3/10/09
Title: Reservations with
costly cancellations in an online game-theoretic model
Abstract: We initiate the
worst-case study of reservations (advance-booking), motivated by the
lack of this feature from current sponsored search systems. We
introduce a simple model for auctioning reservations, in which
impatient, selfish, buyers with private values arrive sequentially and
place a bid. With adversarial arrivals and values, any algorithm can
perform arbitrarily worse than the offline optimum. We show the
effectiveness of a novel, practical assumption: the seller can cancel
at any time an earlier reservation, resulting in a utility loss to the
reservation holder of a fraction of her value. Our main result is an
online algorithm and pricing scheme with many desirable game-theoretic
and optimization properties. Winners have an incentive to be honest.
Bidding one's true value is never worse than any lower bid. Our
mechanism's revenue is within a constant fraction of the a
posteriori revenue of the Vickrey-Clarke-Groves mechanism. Our
mechanism's efficiency is within a constant fraction of the a
posteriori optimally efficient solution. If efficiency also takes into
account the utility losses of buyers whose reservation was canceled,
our mechanism can match an upper bound on the competitive ratio of any
deterministic online algorithm. Based on joint work with Jon Feldman,
Muthu Muthukrishnan and Martin Pal.
Date:
3/13/09
Title: Channel Resource
Management in Multi-channel Wireless Networks
Abstract: Multi-channel
wireless networking is a key enabling technology in the next generation
wireless networks. Through the usage of non-overlapping channel
resources, users can communicate with each other simultaneously over
different channels, and thus the network performance can be
significantly improved by exploring concurrent transmissions. In such a
multi-channel environment, a natural question is how to manage the
channel resources. Relevant issues include channel assignment support
for both unicast and local broadcast, interference mitigation,
throughput optimization, switching delay, overhead reduction, etc. In
this talk, I will first provide a brief overview of the background and
design principles of channel resource management in multi-channel
networking environments. Next, I will address the challenges stated
above and introduce two channel resource management schemes, one for
unicast and the other for local broadcast. I will show how
interference-free communication can be achieved for both unicast and
local broadcast under certain conditions, and how the channel
management scheme can reach 100% throughput for unicast. Additionally,
I will discuss some other related issues such as network scalability
and connectivity. Lastly, I will talk about some future research
directions for this work and how it fits into my overall research
vision.
Date:
3/25/09
Title: Relay Node Placement
in Wireless Sensor Networks for Connectivity and Survivability
Abstract: Wireless Sensor
Networks (WSNs) have been projected as one of the popular networks of
the future. Their infrastructureless nature, ease of deployment, and
relative cost-effectiveness has made them suitable for applications in
both the military and the civilian domains. Despite the inherent
advantages of WSNs, there exist fundamental problems that need to be
solved to expedite their adoption. One such area with extensive
research potential is the design of algorithms for energy aware
routing, improvement in lifetime, and survivability. To prolong network
lifetime while meeting certain network specifications, such as
connectivity and survivability, researchers have proposed to deploy in
a WSN a small number of relay nodes (RNs) whose main function is to
communicate the data from the sensor nodes (SNs) to the base stations
(BSs). Problems in this area are studied under the theme of relay node
placement.
This talk aims to provide a holistic view of the relay node placement
problem. It presents the state of the art for some of the problems and
the existing challenges. The talk deals with single and two-tiered
wireless sensor networks, constrained and unconstrained relay node
placement, and network design to satisfy connectivity and survivability
constraints for the network. It concentrates on the problem of relay
node placement in a single-tiered network setting, under constrained
relay node deployment, for ensuring connectivity and survivability.
Date:
3/27/09
Title: Ontology-driven
Service Discovery on the Web and Artificial Life Approaches on the Grid
Abstract: Two research
projects will be presented during this talk. The first project
addresses one essential requirement of service-oriented environments
which is the ability to locate services of interest. As traditional
approaches to service discovery have generally relied on simple syntax
matches, the proposed approach is based on semantic descriptions stored
in ontologies and reasoning capabilities which are applied to these
semantic descriptions. The approach and the implementation of a
prototype will be presented and discussed.
The second project addresses the problem of load balancing in a Grid
environment. Prior work has addressed load balancing using a
centralized approach, however, as this can lead to a single point of
failure, distributed load balancing using ant colony and particle swarm
approaches have been investigated. The two approaches are implemented
in a simulation environment (GridSim) and different performance
measures (e.g. makespan, load balancing level, etc.) are evaluated.
Date: 3/31/09
Title: Z-SKY: An Efficient
Skyline Query Processor based on Z-order
Abstract: In decision
support applications, we are often interested in finding
optimal data points from a large set of data points in a
multidimensional space. Recently, a skyline query has been proposed to
retrieve those optimal data points, as a subset of data points that are
not dominated by any other point in a database. A data point p is said
to dominate another point q if p is strictly better in at least one
dimension than q and p is not worse than q for all the remaining
dimensions. While data cardinality or dimensionality increases,
efficient evaluation of the skyline query is challenging. Observing a
strong connection between Z-order space filling curve (or Z-order curve
for short) and skyline processing strategies, we design Z-SKY, a query
processor that efficiently supports skyline query processing, skyline
result updates and several skyline query variants. At the core of Z-SKY
is a new index structure called ZBtree that indexes data points based
on
Z-order curve, developed upon which a suite of innovative and efficient
algorithms. All of them are developed based on coherent ideas and
concepts. In this talk, a detailed skyline query analysis, the design
of
Z-SKY as well as the performance evaluation will be presented.
Date: 4/24/09
Title: Web Service
Composition: A Petri Net Approach
Abstract: Web service
framework has evolved to become an important paradigm for distributed
computing. When any single web service fails to accomplish service
requestor’s multiple function requirements, multiple web services need
to be dynamically configured together to form a web service
composition. This work deals with automatic configuration of services
under practical constraints. First, according to the customized or
application-specific web service functional requirement, discover all
the web services. Second, by analyzing function decomposition and
function selection on the service interface information, build a
complete service functional dependency configuration net based on Petri
nets. Third, choose and compute the quality-of-service (QoS) attributes
for the whole configuration. A transformation method is utilized to
change non-linear aggregation functions to linear ones. Relative
importance or value trade-offs of different attributes are represented
through subjective preference or perception. Fourth, the QoS attribute
value for each real web service is gained. An association algorithm
translates and compiles QoS attributes. Finally, the linear programming
problem is formulated and solved. The best configuration is found and
sensitivity analysis is carried out. The concepts and developed
algorithms can be readily put into industrial applications. This talk
will cover some basic knowledge of Petri nets as a fundamentally
important modeling tool for discrete event systems and their particular
applications to modeling, analysis and composition of web services.
Computer Science Seminar Series - Fall 2008
Time 3 - 4PM
|
Date
|
Speaker
|
Title
|
Institution
|
Location
|
9/26/08
|
Elizabeth
Winiarz, Science Librarian
|
Library Resources and Services for Computer
and Information Science - How to Save Yourself A Lot of Time When Doing
Research
|
UMD
Library |
DION 101 |
10/31/08
|
Dr.
Vinod
Vokkarane
|
High-Speed Service-Oriented Optical Internet
|
UMass
Dartmouth
|
DION 101 |
11/21/08
|
Dr. Iren
Valova
|
Role Of Initialization In SOM Networks -
Study
Of Self-Similar Curve Topologies
|
UMass
Dartmouth |
DION 101 |
12/05/08
|
|
Graduate Student Research Presentation
|
UMass
Dartmouth |
DION 101 |
12/12/08
|
Dan Reddy
|
Software Assurance
|
EMC
|
DION 101 |
Date: 10/31/08
Title: High-Speed
Service-Oriented Optical Internet
Abstract: Many emerging
next-generation Internet applications, such as videoconferencing,
multimedia distribution, and high-performance scientific computing, are
characterized by high bandwidth requirements, single-source and
multi-destination communication paradigms, strict QoS requirements with
respect to delay and loss, and high reliability requirements. In
order to support these diverse requirements, emerging networks must be
able to manage resources in a flexible manner. We present the latest
research on next-generation service-oriented optical Internet
architectures and protocols performed at UMass Dartmouth. We will
discuss new services provided at the optical layer and how it will be
beneficial to next-generation Internet applications. The talk
will specifically discuss new algorithms and protocols for implementing
manycast communication service and providing high-speed reliable
communication service.
Speaker Bio: Dr. Vinod
Vokkarane is the Assistant Professor of Computer and Information
Science at the University of Massachusetts Dartmouth. He is the
Director of the Advanced Computer Networks Laboratory (ACN). He
received the B.E. degree with Honors in Computer Science and
Engineering from the University of Mysore, India in 1999, the M.S.
degree in Computer Science from the University of Texas at Dallas
in 2001, and the Ph.D. degree in Computer Science from the University
of Texas at Dallas in 2004. Dr. Vokkarane's research interests include
the design and analysis of architectures and protocols for
next-generation optical and wireless networks. He has been
involved in the organization of several conferences and symposia on
optical networking. He has published over 40 research papers in
refereed journals and conferences, including a paper that
received the Best Paper Award at IEEE GLOBECOM 2005. He is also a
recipient of the Texas Telecommunication Engineering Consortium
Fellowship 2002-03, and the University of Texas at Dallas Best
Dissertation Award 2003-04. Dr. Vokkarane is the co-author of a
book, "Optical Burst Switched Networks," Springer, 2005. Dr.
Vokkarane has received research grants from National Science Foundation
(NSF), United States Marines Corps (USMC), and regional networking
companies.
Date: 11/21/08
Title: ROLE OF
INITIALIZATION
IN SOM NETWORKS - STUDY OF SELF-SIMILAR CURVE TOPOLOGIES
Abstract: This work
investigates the initialization process in SOM. This is of importance
because of the issue of network linearity and, subsequently, the
quality of the produced map. We discuss 1D classical SOM, i.e. the
algorithm presented by Kohonen, and experiment with three different
approaches to initialization - random, random with training or priming,
and using self-similar curves to initially position the neurons. Our
results show that, while the network will eventually untangle when
random initialization is used, this will occur at the 100,000+ epoch.
With priming or self-similar curves, the final, linear map is produced
much earlier, i.e. 10,000th epoch at the most. The benefits of this are
obvious with significantly reduced time to produce a usable map of the
input space.
Date: 12/12/08
Title: Software Assurance
Abstract: Software
Assurance: As Customers (especially governments) buy Commercial Off the
Shelf Software (COTS) what level of assurance should
they expect regarding how secure the software is as built and
implemented ? This session will provide an overview of what
exists today and what trends are emerging related to software assurance
from a government and industry perspective. Included will be a
discussion of how these trends could influence the Systems Engineering
or IT skills needed in the field and new employment opportunities that
may exist.
Speaker Bio: Dan Reddy is a
Consulting Product Manager in the Product Security Office at EMC, a
group that is charged with the continued driving of security
improvements into EMC products. In his various roles within his 12
years at EMC he has been consulting with EMC Customers around product
security issues and has been involved in numerous IT software
development projects. Prior to joining EMC, Dan spent 15 years at New
England Electric, a
major electric utility with nationally critical infrastructure where
held a variety of IT and business roles including Manager of Technical
Services in IT and Staff Assistant to the Chief Operating Officer. He
also teaches Computer Science courses at Quinsigamond Community College
in Massachusetts where has taught for over 30 years. He holds and M.
Ed. in Computer Science from Worcester State College and a B.A. from
Tufts University in Education.