Page 33 - profile2012.indd
P. 33
Research Laboratories 研究群
網路系統與服務實驗室
Network System
and Service Laboratory
Research Faculty
Research Faculty
Jan-Ming Ho Ling-Jyh Chen MengChang Chen Sheng-Wei Chen Wen-Tsuen Chen Tyng-Ruey Chuang Jane W. S. Liu
Research Fellow Associate Research Fellow Research Fellow Associate Research Fellow Distinguished Research Fellow Associate Research Fellow Chair Research Fellow
Group Profile
In this Laboratory, our research address several aspects of net- algorithms to support spatio-temporal data calculation using use of data and information from inde-
work systems and services, including improving wireless and the compressed data directly. The results of our research have pendent sources by independently de-
delay-tolerant network protocols, leveraging human computa- been implemented in two real-world networked sensing sys- veloped applications and services and
tion capabilities to address key challenges, developing critically tems: One is a mission-critical sensor network, called Yushan- can readily accommodate new informa-
needed information and communication technologies for dis- Net, for hiker tracking, search, and rescuing in Yushan National tion sources and applications as needed Making the world a better
aster management, and seeking solutions to network compu- Park, and the other is a mobile phone sensing system, called in response to unforeseen crisis situa-
tational problems on providing large-scale financial risk man- TPE-CMS, for comfort measurement of public transportation tions. Our work is now supported by an place with innovations in
agement services. systems in the Taipei metropolitan area. Academia Sinica thematic project called
OpenISDM (Open Information Systems for Disaster Management). The cur- network, system and service
Our work on wireless communication and networking focuses Applications of human computation range from the exploita- rent works by members of our laboratory include the development of smart
on multi-hop wireless communication, mobility management tion of unsolicited user contributions, such as using tags to aid cyber-physical devices and applications as elements of disaster-prepared technologies.
and network forensics. We are developing techniques to pro- understanding of the visual content of yet-unseen images, to smart environment; strategies for crowdsourcing collection of sensor infor-
vide end-to-end quality of service (QoS) guarantees for com- utilizing crowdsourcing platforms and marketplaces (e.g., Ama- mation to complement of data from in-situ physical sensors; methods and
munication flows in wireless mesh networks. We are investigat- zon’s Mechanical Turk) to micro-outsource to a large population tools for reducing human efforts in collecting, validating and refining disas-
ing several mobility and handover management schemes that of workers tasks such as semantic video annotation. Further, ter related information in social reports; methods and tools supporting com-
support service at anytime and anywhere and have proposed crowdsourcing offers a time- and resource-efficient method for munication and computation infrastructures for gathering, caching, fusion
architecture and protocols that support seamless handover for collecting large input for system design or evaluation purposes. and distribution of ubiquitous and heterogeneous real-time streaming sen-
both terminal and network mobility. We also study the problem We are applying crowdsourcing to optimize computer systems sor data and information to response centers and individual responders and
of detecting a slow attack by modeling outbound connection more rapidly and to address human factors more effectively. In volunteers during disasters; the exploitation of complementary merits of dif-
attempts of a host as a time series and use spectral analysis to the past few years, we have performed extensive studies on the ferent network access technologies, approaches, and network types to make
discover recurrent events generated by the potential attack performance of GWAP (Games with A Purpose) systems and de- the physical connectivity as robust as possible during and after disasters.
amid legitimate traffic. The regularity of these attack attempts signed a human computation game in order to collect diverse
is preserved in the time series and can be observed in the fre- user annotations efficiently. In addition, we have proposed a We are also parsuevy in studying network computation problems on provid-
quency domain. cheat-proof framework that can be used to assess the quality ing large-scale risk management services. Despite of the long history in the
of experience provided by multimedia content effectively. We development of economic and financial theories and practices in financial
In addition, we study networked sensing systems with focuses have found that crowdsourcing is indeed a powerful strategy risk management, the “worldwide credit crisis” in 2008 had manifested the
on energy efficiency and large-scale sensor data management. that can draw collective intelligence for AI-hard problems. In vulnerability of the current financial industry in tolerating risks. Several ex-
We have developed an adaptive GPS scheduling algorithm to the future, we will continue to study how to use crowdsourcing amples show that even the three major rating agencies were unable to re-
prolong the lifespan of GPS-enabled mobile sensors, and de- well to overcome challenges in a variety of areas. port major default events efficiently. In Enron’s bankruptcy case in 2001, its
signed a context-aware duty cycle algorithm to adjust duty bonds had maintained “investment grade” ratings until five days before the
cycles of GPS receivers and radio interfaces of mobile sensor A disaster management information system (DMIS) facilitates company declared bankruptcy. In Lehman Brother’s case in 2008, they still
devices in accordance with surrounding contexts inferred by the access, use and presentation of data and information by received “investment grade” ratings on the morning they declared bankrupt-
low-cost sensors, application systems and services that support decisions and cy. The rating companies claim that their rating reports provide a long-term
e.g., accelerometers operations during all phases of disaster management. State-of- perspective rather than providing an up-to-minute assessment. In rating
and thermometers. the-art systems have several common limitations: They cannot credit of a company, there are hundreds of firm-specific and macroeconomic
We have proposed make good use of information sources owned by businesses, variables. There is no doubt that assessing credit risk in real-time is indeed
a lightweight and organization, communities, and so on during emergencies; do a task with high computational complexity. Nevertheless, it is an important
lossless data com- not exploit synergistically information from networks of things foundation of maintaining stability of the financial market. Complementary
pression algorithm and crowd of people and are not sufficiently agile in response to the research in economic and financial theories and practices in risk man-
for spatio-temporal to changes in disaster situation. We are collaborating with re- agement, we aim at developing computing technologies based on cloud
data, and designed searchers in Institute of Earth Science and Center for Climate computing framework towards provisioning of large-scale real-time financial
a set of data query Changes and Computer Science and Engineering faculty mem- risk management services, including (1) real-time rating of company credit;
bers from several leading universities in Taiwan and USA to de- (2) real-time rating of personal credits; and (3) pricing and risk measures of
velop an open framework for DMIS free of these limitations. A financial products.
system built within the framework can support the access and
研究群
32 Research Laboratories
32
33
33