Page 34 - profile-ok
P. 34
研究群 | Research Laboratories
Network System and Service
Laboratory
tion and reconnection are frequent, and link performance is highly Network coding Network coding can increase the throughput of a cal framework, build theoretical models, and conduct lab and field ex- correlation between the OBD-II sensing data of vehicles and the G-
variable and poor. We tackle this problem from both the network- capacitated network with multicast. Most works on network coding periments. Our research covers the following topics: contract design sensor readings, we model users’ driving behavior and infer their
ing side and the sensing side. Regarding the networking aspect, we typically leverage network coding to solve various network prob- and clearing rules, trading mechanisms, pricing, multi-dimensional, eco-friendliness. Using emerging social network platforms, we allow
model and analyze network mobility and design effective schemes lems. The resulting solutions are not viable, however, because of their conditional contracts, contract clearing methods for preference con- users to share their driving experience, compare their carbon foot-
for network data dissemination. Regarding the sensing aspect, we in- incompatibility with current Internet standards and routers. Our re- tracts, trader information, trader behavior and market efficiency, trad- prints, find the most comfortable route, and learn how to drive more
vestigate fault-tolerant mechanisms to facilitate the reliable harvest- search regards network coding in a different perspective. Our objec- er risk attitude and behavioral analysis, market manipulations, price eco-friendly with their friends. Moreover, by analysis and statistics of
ing of sensing data in challenged environments, and develop energy tive is to leverage the user access patterns in a client-server model to spreads and market efficiency, correlation between market prices the collected data, the VProbe can identify the most uncomfortable
efficiency mechanisms to prolong the lifespan of sensor nodes. minimize the resource consumption in the server with network cod- and news events, and trader geo-location analysis. road segments, find most energy-saving route, and detect abnormal
ing. We proved that the problem of finding optimal network coding is driving behavior. Therefore, it shows premises in facilitating transpor-
In addition to theoretical study, we prototyped our research results NP-hard in this case and then designed an approximation algorithm ● Open Geospatial Information Processing tation planning, route planning, traffic event detection, and future
and developed real-world systems to provide networked sensing with guaranteed performance bounds for all instances. We also ad- cyber-physical social network systems.
services. As an example, we have implemented a delay-tolerant sens- dressed the decoding cost, and devised an algorithm that obtained Geo-information is abundant on the Web: place names, GPS traces,
ing system, called YushanNet, for mountaineering applications. The the optimal decoding cost for clients. Our algorithm requires no mod- maps, and photos, just to name a few examples. As web services are 4. Quality of Experience
system employs light-weight devices with low-power, short-range ra- ification of Internet standards and existing routers. increasingly driven by user-generated content and are often created
dio, which then exploit opportunistic hiker encounters to record and on end-user devices, we are dealing with more and more user-gener- In the design of computer systems, user satisfaction is an important
disseminate hikers’ whereabouts in the mountains. The system has 3. Social Computing Research ated geospatial information. Our research focuses on the use of se- factor (if not the most important one) in determining the merits of
been approved to be deployed on the Yushan Peak Trail, and it will mantic web technologies, web standards, and open source software a system. No matter how good the system-level performance met-
soon become a regular service of the national park to provide hiker Our research in the area of social computing includes prediction mar- tools in building new geospatial applications. We emphasize the use rics of a system are, the system cannot be considered good from its
tracking, environment monitoring, and other advanced e-services to kets, open geospatial information processing, and human computa- of open standards, such as RDF/OWL, GML/KML, and SVG, for repre- end-users’ point of view if its users feel dissatisfied or frustrated when
tourists. tion and social sensing applications. senting geospatial knowledge in ways that are easy to discuss, reuse, using the system. However, user satisfaction is an abstract feeling
and remix.
that cannot be observed or measured directly. For this reason, how
● Multicasting ● Prediction markets Currently we are investigating research possibilities in folk knowl- to ensure that a computer system will satisfy users under all kinds of
Distributed resource allocation of heterogeneous wireless networks Information markets are an emerging research area for information edge about places (i.e., places where peoples live, or once lived) and conditions is still an open problem. The problem can be divided to
three sub-problems: 1) how to measure user satisfaction systemati-
Although finding the shortest path in a distributed manner is achiev- retrieval and the aggregation of interactive trading. Participants the associated information technologies needed for the processing cally and efficiently; 2) how to model the relationship between user
able in polynomial time, multicast routing in heterogeneous wireless of information markets trade for the outcomes of future events, of- of such knowledge. We are especially interested in geospatial knowl- satisfaction and system-level performance metrics; and 3) how to ad-
networks is a difficult problem because it becomes NP-hard in this ten called contracts, a term borrowed from financial markets. The edge about aboriginal settlements, and are developing the necessary just system parameters systematically to achieve the desired tradeoff
case. Here, we devised a distributed algorithm based on optimization clearing prices of contracts are decided by the actual outcomes of ontologies to help collect and integrate information sources related along multiple dimensions of user satisfaction.
techniques and derived a lower bound on the optimal solution. Our underlining events, or by the intentions of the majority participants to aboriginal settlement migrations in Taiwan. As such, this is an inter-
algorithm enables more mobile hosts to cluster together and leads to if the events do not have any definite outcomes (such as “What are disciplinary project involving Taiwan aboriginal studies, geospatial We have proposed a novel experimental framework that can solve
the use of fewer cells to save scarce wireless bandwidth than those the most influential techniques in 2009?”). Information markets have data, ethnographies and gazetteers, ontology, knowledge represen- the first problem. By using paired comparison and an algorithm to
appeared in literature. Additionally, the paths in the multicast tree been used to predict elections, movie box offices, product sales, tation and reasoning, and Web technologies in general. verify the consistency of each participant’s input, the proposed
connecting to the selected cells share more common links, helping project schedules, macroeconomic indicators, and even international framework reduces the difficulty of obtaining quality judgments,
save wire-line bandwidth. The algorithm can be implemented in mo- political and economic risks. Many multinational corporations, such ● Human Computation and enables researchers to invite Internet users to participate in their
bile handsets, and requires no modification for use with current In- as Google, Microsoft, Intel, Hewlett-Packard, Archelor and BP-Amaco, The rationale behind human computation is to outsource certain quality assessment experiments. The evaluation results demonstrate
ternet standards. It is therefore compatible with existing commercial have started to apply the techniques of information markets for fore- steps of the computational process to humans and let them solve the that the proposed framework enables researchers to outsource their
products and can be directly implemented on mobile handsets for casting and monitoring project schedules, facilitating new product problems that are intuitive to humans, but that computer technolo- experiments to an Internet crowd without risking the quality of the
ISPs. development, and managing demand risk. Preliminary results in this gies cannot solve completely. Examples of such problems are image results, while at the same time obtaining a higher level of participant
area are quite promising, because the predictive accuracy of informa- diversity at a lower monetary cost.
Multicast network planning of LEO satellite network. The problem of tion markets continues to exceed traditional opinion polls. Scholars annotation and commonsense reasoning.
multicast network planning in LEO satellite networks is to find a Recti- of politics, economics, sociology, psychology, finance, public policy, Among various human computation systems, we tackle `Games with To solve the second question, we proposed to measure user satisfac-
linear Steiner Tree (RST) for each multicast stream. We took advantage and information/computer science all began to investigate its theo- a Purpose’ (GWAP) systems. These systems take advantage of people’s tion based on the large-scale, passive measurement of user behavior
of the fact that each LEO satellite network includes at most a few hun- rems, models, and empirical evidence. desire to be entertained, and produce useful metadata as a by-prod- in real life. Our method models the relationship between user satis-
dred satellites and proposed a novel integer linear programming for- uct. We argue that in order to collect human intelligence more effi- faction and system-level performance metrics, such as network de-
mulation. Based on this formulation, our solution is at least 10 times This project focuses on the efficiency and efficacy of information ciently, GWAP systems must be designed and played with strategies. lay, processing delay, and bandwidth, by statistical methods. Because
faster than existing solutions within the literature. Multicast streams markets in both controlled and open environments. Specifically, we We developed a generic analysis framework to investigate intrinsic it enables arithmetic computation and even the prediction of user
are usually dynamically added to networks for new applications and approach the study of information markets from the perspective of properties of GWAP systems, and conducted a set of design strategies satisfaction, it opens up a novel approach for user satisfaction mod-
services. Having a 10 times faster solution means significant reduc- information science, and emphasize the applicability of information to improve the performance of GWAP systems in terms of efficiency eling. In addition, we explored the third question (i.e., how to adjust
tion in network planning time whenever new streams are added. markets in event predictions and market preference surveys based on and quality. Moreover, we have implemented and verified the de- system parameters systematically to maintain user satisfaction). We
information retrieval and aggregation. We aim to develop a theoreti-
signed strategies in real-world GWAP systems and have released the took Voice over IP (Internet Telephony) as an example, and proposed
source code and game traces to the research community. an efficient algorithm based on a statistical regression approach to
adjust multi-dimensional system parameters. The reported results
● Social Sensing Applications demonstrate that the proposed algorithm can provide users with
Our research on social sensing systems aims to exploit the synergic good VoIP experience in all cases by adjusting the system parameters
use of sensing systems and social networks. A device that integrates automatically.
sensing systems and social networks is in essence a cyber-physical
social network system.
For the sake of proof of concept, we implement the VProbe system
that employs modern smart phones to collect GPS trajectories and
G-sensor readings in users’ daily transportation. By investigating the
34 35