ITIM 2011
ITIM 2011 International Conference
Beijing, China | March 20 to Marth 24, 2011
The Third Symposium on Intelligent Technologies and Information Management for Knowledge Society
Organized by
State Key Laboratory of Intelligent Technology and Systems,
Supperted by
Department of Computer Science and Technology,
Graduate
Invited Speakers
Workshop Program
21 March 2011 (Monday), 1st Day, meeting room: FIT 1-315
9:00 - Registration
Opening Address
9:30-9:40 Professor Maosong Sun (Tsinghua University)
Keynote Speeches
9:40-10:10 Professor Bo Zhang (Tsinghua University)
10:30-11:00 Poster & Coffee Break (lounge)
Invited talks
10:30-11:00 Professor Yasuyuki SUMI (Kyoto University)
Experience Medium: Toward a New Medium for Exchanging Experience
11:00-11:30 Professor Maosong Sun ( Tsinghua University)
11:30-12:00 Professor Shigeo MATSUBARA (Kyoto University)
Incentive Design toward Fostering Collective Intelligence
12:00-12:30 Professor Xiaoyan Zhu (Tsinghua University)
12:30-14:00 Conference Lunch (Yanmingyuan Restaurant)
14:00-14:20 Professor Masayuki MUKUNOKI( Kyoto University)
Sensing Web Project
14:20-14:40 Professor Daisuke Kawahara( Kyoto University)
Knowledge-oriented Natural Language Processing for Information Analysis
14:40-15:00 Professor Minlie Huang (Tsinghua University)
15:20-15:20 Professor Satoshi NAKAMURA( Kyoto University)
Reranking Search Results Based on Simple Interaction and Social Annotation
15:20-15:40 Professor Yasuhito ASANO( Kyoto University)
Enishi: Relation Search Utilizing Wikipedia and the Web Complementarily
15:40-16:00 Professor Tetsuya Ogata (Kyoto University)
Constructivist Approach to Human-Robot Interaction from Dynamical Systems Perspective
16:00-16:30 Coffee break (lounge)
16:30-16:50 Professor Min Zhang (Tsinghua University)
16:50-17:10 Professor Jie Tang (Tsinghua University)
17:10-17:30 Professor Lifeng Sun (Tsinghua University)
17:30-17:50 Professor Jianyong Wang (Tsinghua University)
18:00- Conference Dinner (Peking Duck Restaurant)
Speakers' Informations
Prof. Yasuyuki SUMI
Graduate School of Informatics (Department of Intelligent Science and Technology)
Homepage:http://www.ii.ist.i.kyoto-u.ac.jp/~sumi/
Experience Medium: Toward a New Medium for Exchanging Experiences
In this talk, I will present a notion of "experience medium" in which we can exchange our experiences in museum touring, daily meetings, collaborative work, etc. The experience medium is a medium for capturing, interpreting, and creating our experiences, i.e., not only verbalized representations of our experiences but also their contextual information (awareness, common sense, atmosphere). I will show our previous and ongoing projects as follows: Building a context-aware mobile assistant for guiding museum visitors and facilitating communications among the users by casual chats between the users' guide characters and comic-like diaries based on their visiting records; Collaborative capturing and interpretation of experiences like conversations, staying together, and gazing something by ubiquitous and wearable sensors; and Supporting systems of casual communications by facilitating to share photos and comments among community members.
BIO
Yasuyuki Sumi is an associate professor in the Graduate School of Informatics at Kyoto University since 2003. Before joining Kyoto University, he had been a senior researcher at ATR for eight years. He received his B.Eng. degree from Waseda University in 1990, and M.Eng. and D.Eng. degrees in information engineering from the University of Tokyo in 1992 and 1995, respectively. His research interests include knowledge-based systems, creativity supporting systems, interface/social agents, ubiquitous/wearable computing, Web intelligence, multimedia processing, and their application for facilitating human interaction and collaboration. From 1997 to 2002, he led a project to build a personal agent system for guiding its users according to their contexts and facilitating their communications during tours in museums, conferences, trade shows, etc. From 2002, he started a project to build a so-called experience medium to capture, understand, and facilitate user experiences by mobile, wearable, and ubiquitous systems. Recently, he has been emphasizing to build a hardware/software environment for measurement, analyzing and supporting of human conversational interactions in order to realize the dictionary and grammar of human nonverbal behaviors.
Prof. Shigeo MATSUBARA
Graduate School of Informatics (Department of Social Informatics)
Homepage:http://ai.soc.i.kyoto-u.ac.jp/~matsubara
matsubara@i.kyoto-u.ac.jp
Incentive Design toward Fostering Collective Intelligence
How to fostering collective intelligence and better collective decision-making is an important issue. Internet auction is an example of collective decision-making, where winners and the payments are determined. So far, auction had been viewed as a competitive place. However, if we can learn the latent cooperation structure that a seller’s behavior benefits another seller, it enables us to design a more efficient auction market. In this talk, I show the result of analyzing the buyout price in Yahoo! Japan auction. Recently, Internet auction sites offer a buyout option. If a seller chooses to use a buyout option, shoppers can purchase items at a set price even when the merchandise is also listed in an auction. In auction sites, identical goods are sold in an auction with and without a buyout option simultaneously. To investigate the interaction between these two selling formats, we have analyzed the real auction data obtained from Yahoo! JAPAN auction, characterized the typical strategies of the seller, built a model, and carried out a simulation on it. The result of the simulation on the model shows that if the probability that a buyer is risk averse is quite high, both two sellers can benefit by selling a good by using a buyout option and selling another good by using an ascending auction.
BIO
Shigeo Matsubara is an associate professor of Department of Social Informatics, Kyoto University. From 1992 to 2006, he was a research scientist of NTT Communication Science Laboratories, NTT. He received his Ph.D. degree in Informatics from Kyoto University. During 2002-2003, he was a visiting researcher at University of California, Berkeley. He was also an advisor of NICT Language Grid project from 2006 to 2007. His research focuses on multiagent systems and information economics. He has published in Artificial Intelligence Journal and other academic journals. He has served as a PC member for international conferences including AAMAS, IJCAI, AAAI and an industrial track co-chair of AAMAS2007.
Prof. Masayuki MUKUNOKI
Graduate School of Informatics (Department of Intelligent Science and Technology)
Sensing Web Project
In this talk, I will present the achievements of the Sensing Web project, launched in the fall of 2007 and finished in the spring of 2010 in Japan. The project's aim is to open the data obtained by the sensors existing in our daily living environment for various purposes. Since the sensor data provide real world information in real-time, we can know various kinds of activities occurring in the real world through those data. We transform the sensor data into this new worldwide social information infrastructure, called the ``Sensing Web'', and developed the technologies for it. In this talk, we overview the technologies emerging from our project and show the results of demonstration experiments in this project.
BIO
Masayuki Mukunoki received the BS, MS and DE degrees in information engineering in 1991, 1993, and 1999, respectively, from Kyoto University. From 1996 to 2002, he was a Research Associate in Kyoto University. From 2002 to 2009, he was an Associate Professor in Hiroshima City University. Since 2009, he has been an Associate Professor in the Accademic Center for Computing and Media Studies and a faculty member in the Graduate School of Informatics, in Kyoto University. His research interests include video media processing, lecture video analysis and human activity sensing with camera.
Prof. Daisuke Kawahara
Department of Intelligence Science and Technology, Graduate School of Informatics
Homepage:http://nlp.kuee.kyoto-u.ac.jp/~kawahara/
dk@i.kyoto-u.ac.jp
Knowledge-oriented Natural Language Processing for Information Analysis
A vast amount of information has been accumulated and circulated on the Web. This situation provides us plentiful knowledge sources, from which we can acquire linguistic knowledge to benefit to basic analysis tools and applications of Natural Language Processing (NLP). In this talk, I will introduce our work on knowledge acquisition from the Web, and also describe our NLP analyzers based on the acquired knowledge. Then, I will demonstrate an NLP application based on these analyzers and knowledge, called WISDOM. WISDOM is an information analysis system that organizes and overviews Web information on a given topic. Our evaluation indicates that this system is effective in people's critical thinking and decision making in their daily lives.
BIO
Daisuke Kawahara received his B.S. and M.S. in Electronic Science and Engineering from Kyoto University in 1997 and 1999, respectively. He obtained his Ph.D. in Informatics from Kyoto University in 2005. He is currently an associate professor of the Graduate School of Informatics at Kyoto University. His research interests center on natural language processing, in particular knowledge acquisition and text understanding.
Prof. Satoshi NAKAMURA
Graduate School of Informatics (Department of Social Informatics)
Homepage:http://snakamura.org/
nakamura@dl.kuis.kyoto-u.ac.jp
Reranking Search Results Based on Simple Interaction and Social Annotation
It is not easy for users to find appropriate contents in any search services and to inform their search intention to search services because interactions between users and search services are limited. In this talk, I will present some approaches for users to find target contents by using additional simple interaction and social annotations such as social bookmarks, social tagging, and comments posted by many users. In this talk, I will present and demonstrate our previous and ongoing projects as follows: Reranking any search results in any search services only by simple interaction; Reranking Web search results by a sense; A ranking method of Web search results by utilizing social bookmarks; and A ranking method of video clips based on impression extracted from social comments.
BIO
Satoshi Nakamura is an associate professor in the Graduate School of Informatics at Kyoto University since 2009. He received the B.E. in Information Systems, Department of Information Systems Engineering from Osaka University, in 1999. He received the M.E. and PhD degrees in Information Systems, Department of Information Systems Engineering, Graduate School of Engineering, Osaka University, in 2001 and 2004. He worked as a research fellow at the National Institute of Information and Communications Technology from 2004 to 2006. He worked as an assistant professor at Kyoto University from 2006 to 2007. Since 2009 he has been an associate professor at Kyoto University. His research interest includes human computer interaction, social annotation, search interaction and so on.
Prof. Yasuhito ASANO
Graduate School of Informatics (Department of Social Informatics)
asano@i.kyoto-u.ac.jp
Enishi: Relation Search Utilizing Wikipedia and the Web Complementarily
How global warming and agriculture mutually influence each other? It is possible to answer the question by searching knowledge about the relation between global warming and agriculture. As exemplified by this question, strong demands exist for searching relations between objects. However, methods or systems for searching relations are not well studied. In this talk, we introduce our relation search system named “Enishi.” Enishi supplies a wealth of diverse multimedia information for deep understanding of relations between two objects by complementarily utilizing knowledge fromWikipedia and theWeb. Enishi first mines objects explaining a relation between two objects from Wikipedia by utlizing our generalized flow model. Enishi then search more multimedia information about relations on the Web using the objects.
BIO
Yasuhito Asano received B.S., M.S. and D.S. in Information Science, the University of Tokyo in 1998, 2000, and 2003, respectively. In 2003-2005, he was a research associate of Graduate School of Information Sciences, Tohoku University. In 2006-2007, he was an assistant professor of Department of Information Sciences, Tokyo Denki University. He joined Kyoto University in 2008, and he is currently an associate professor of Departmento of Social Informatics, Graduate School of Informatics. His research interests include web mining, network algorithms. He is a member of IEICE, IEEE, IPSJ, DBSJ, OR Soc. Japan.
Prof. Tetsuya OGATA
Graduate School of Informatics (Department of Intelligence Science and Technology)
Homepage:http://winnie.kuis.kyoto-u.ac.jp/~ogata/hp
ogata@i.kyoto-u.ac.jp
Constructivist Approach to Human-Robot Interaction from Dynamical Systems Perspective
In human robot interaction, robot should understand (ground) the real environment to “symbols” which dynamically changes with interaction context. This talk introduces my synthetic approach for designing human robot interactions as dynamical systems. This approach couples multiple dynamics of the neuro-dynamical system, the real-robot system, and the environment to describe and/or emerge explicit symbols using dynamical systems form. The neuro-dynamical system consists of multiple recurrent neural networks of which time scale is different, and each neuron group self-organizes the dynamical representation of the robot’s sensory-motor flow. In this talk, I present our recent studies on infant model of motion imitation, phoneme acquisition, multi-modal translation, and motion-language translation.
BIO
Tetsuya Ogata received the BS, MS and DE degrees in Mechanical Engineering in 1993, 1995, and 2000, respectively, from Waseda University. From 1999 to 2001, he was a Research Associate in Waseda University. From 2001 to 2003, he was a Research Scientist in the Brain Science Institute, RIKEN. Since 2003, he has been a Faculty Member in the Graduate School of Informatics, Kyoto University, where he is currently an Associate Professor. Since 2005, he has been a Visiting Associate Professor of the Humanoid Robotics Institute of Waseda University. Since 2009, he has been a researcher at JST PRESTO (5 years). His research interests include human–robot vocal-sound interaction, dynamics of human–robot mutual adaptation and active sensing with robot systems.
Prof. Maosong SUN
Department of Computer Science and Technology in Tsinghua University
Homepage:http://nlp.csai.tsinghua.edu.cn
I got B.E, M.E.degree in Tsinghua University in1986 and 1988.And I received Ph.D. in Computational Linguistics of City University of Hong Kong in 2004.
My research interests are computational linguistics, statistical and corpus-based natural language processing (NLP), Chinese language computing (computational morphology, bilingual terminology extraction), information retrieval (Chinese text categorization, graphical model-based keyword extraction), collective intelligence (tag generation, Web trend analysis) and social computing (query log analysis, community discovery). I have participated as project leader or principal researcher in over 20 projects funded by National Natural Science Foundation of China, National Social Science Foundation of China, National 863 High-Tech Program, National 973 Basic Research Program, as well as in projects funded by a number of international IT companies. I have published, together with my students, about 130 papers in academic journals and international conferences in the above fields. The total number of citations of these papers in Google Scholar is roughly 1,400. I have served as program committee member in numerous national and international conferences, and as conference chair or program committee chair for many times.
One of my research focus is Chinese word segmentation, the most fundamental issue in Chinese information processing. I have proposed some key concepts in word segmentation (such as maximal overlapping segmentation ambiguities, true and pseudo segmentation ambiguities, local and global statistics), and have developed an integrated Chinese word segmentation and part-of-speech tagging system, which is able to explore all sorts of knowledge, bigrams of words, parts-of-speech and characters, statistical and structural information of named entities, and local statistics of character strings. I have also tried to extend my experience in Chinese word segmentation to other languages in which word segmentation are need, resulting in an international standard "ISO/FDIS 24614-1: Language Resource Management -- Word Segmentation of Written Texts -- Part 1: Basic Concepts and General Principles". I am the only project leader for this ISO standard.
Recently, I have presented an original viewpoint in NLP: NLP based on huge-scale naturally annotated corpora. The basic idea is with Web-scale corpora, natural annotation may help machine better perform some NLP tasks. There are two types of natural annotation: explicit (such as punctuations, anchor text, query log, Wikipedia, blog tags) and implicit (such as language usage patterns). I further put forward a fundamental problem: if we could integrate all information drawn from naturally annotated corpora of different perspectives, are we able to achieve some degree of deep understanding of languages A preliminary work by my students and me published in Computational Linguistics 2009 showed the usefulness of punctuations in Chinese word segmentation, suggesting that this idea deserves further study.
Prof.Shiqiang YANG
Department of Computer Science and Technology in Tsinghua University
Homepage:http://media.cs.tsinghua.edu.cn/~yangsq
I got Bachelor and Master degree in Tsinghua University in 1977 and 1983.
My research group addressed multimedia information analysis in the early years, covering most of the key techniques including video/image compression and codec, media streaming and transmission, distributed multimedia system, content-based image/video retrieval, etc. With the rapid development of social media, we integrate Internet users' behavior with web information, and address the issue of user modeling in social network and the diffusion mechanism of social media in social network.
In recent years, we've undertaken a lot of national key research projects from NSFC, National Basic Research Program of China (973), National 863 High-Tech Program, etc. We've published more than 100 research papers. Some of their 30+ alumni have received Excellent Ph.D. Dissertation Award from China Computer Federation (CCF). In addition, several research papers have been awarded or nominated as Best Paper by international transactions and conferences, including IEEE T-CSVT, ICASSP, GlobalComm, etc. Among them, the heuristic scheduling algorithm they proposed for media processing on multi-core processor architecture is awarded as the Best Paper of IEEE T-CSVT in 2009.
The research achievements of their group have been widely adopted in various fields and industries. Their P2P streaming media system-GridMdia-was adopted by China Central TV (CCTV) in 2004, which is one of the largest TV networks in the world.
Prof. Xiaoyan ZHU
Department of Computer Science and Technology
Homepage:http://www.csai.tsinghua.edu.cn/~zxy/
I got Bachelor degree in University of Science and Technology Beijing in 1981.And I received Master of Automation in Kobe University in 1989, got Ph.D. in Nagoya Institute of Technology in Japan in 1990.
I am interested in a wide range of research topics in intelligent information processing. Before 2004, my group focused on Optical Character Recognition (OCR), Speech Signal Processing and Human-Computer Interaction. The theoretical achievements of our work have been applied to two systems-OCR Engine (a system for handwritten character recognition) and Aurora (a system developed for blind and visually impaired computer users). Thanks to our OCR Engine's high accuracy, it had been used in the Fifth National Census of China in 2000. The Aurora system provides users with a set of helpful tools including screen reader, speech recognition server, Braille translator, and email editor. Aurora has a large population of organizational (schools for the blind) and individual users in China, and was adopted by China's First Computer Certification Test for the Blind. Starting 2004, our work has been focusing on Text Mining-Biomedical Text mining and Question Answering (QA) in particular.
Information distance is a fundamental concept in our theoretical research. To measure informational similarity between individual objects, they proposed conditional information distance model, MIN information distance model and multiple-object information distance model, respectively. These models extend the traditional information distance theory, and provide creative and effective ways to practical application. We have successfully applied these models to many text mining tasks such as pattern optimization in biomedical text mining, information distance between a question and its answer in QA, content similarity measurement in multiple document summarization, and typical/overall document extraction in product review summarization. Our multiple-document summarization system successively ranked 1st in TAC (Text Analysis Conference) 2008 and 2009, a well-recognized international evaluation for Natural Language Processing methodologies.
Improving user experience in internet information acquisition is the long-term goal of our research. Based our research results in text mining, we have developed online systems to facilitate users in acquiring domain-specific or general information from the internet. These systems include ONBIRES (Ontology-Based Biological Relation Extraction System), QUANTA (Question Answering System), and PROCAR (Product Review Mining System). Our bioinformatics text mining system ranked 1st in several tests of BioCreative II-a critical assessment of information extraction systems in bioinformatics domain, and has attracted the attentions of NCBI (United States National Center for Biotechnology Information)-a prestigious institute which provides biomedical databases and research tools for scientific users. Beginning in 2008, our collaboration with NCBI focuses on the areas of function summarization, and gene/protein name entity identification. At present, NCBI is planning to apply our biomedical texting mining techniques in its biomedical literature search engine named PubMed. Fairly soon, our techniques will serve world-wide biomedical scientific users through PubMed. Our Question-Answering (QA) system aimed at providing users with direct answers to any general questions (e.g. weather report, stock quote or currency exchange). These answers are extracted from the internet. Our QA system can be customized to accommodate different information sources, forming search engine-based QA, encyclopedia-based QA and community-based QA, respectively. This promotes its efficiency while retaining generality, and is most useful in a mobile scenario. We expect our internet information acquisition service would be available to anyone, in anytime and at anywhere.
My internet information acquisition research is currently supported by Canada's International Development Research Centre (IDRC). IRDC provided us one million Canadian dollars as financial support, and appointed me as the IDRC Research Chair in Information Technology to conduct this research.
Prof.Jie TANG
Department of Computer Science and Technology
Homepage: http:// /keg.cs.tsinghua.edu.cn/persons/tj
jietang@tsinghua.edu.cn
I got Ph.D. in Computer Science & Technology of Tsinghua University in 2006.
My research group addresses fundamental problems in social network analysis, data mining, and semantic web. In particular, I focus on modeling social network, tracking users' behaviors, and understanding the evolution of network structures. I have been a visiting scholar at University of Illinois at Urbana-Champaign, the Chinese University of Hong Kong, Hong Kong University of Science and Technology, and Leuven University. I am the principle investigator of several national projects and international joint fundings.
My current research topics mainly include: 1) Semantic information extraction and integration on the Web. We have proposed various methods for semantic information extraction, which are currently successfully used in several practical applications. 2) Topic-level ranking over heterogeneous networks. In contrast to traditional keyword-based search and object-oriented search, our topic-level search aims at understanding the underlying semantics of documents and queries, and matching them at the topic level. 3) Social action tracking and influence analysis. We have proposed a method to gain insight into the micro-level dynamics of social networks such as how an individual user influences his/her behaviors (actions). To the best of our knowledge, we are the first to quantify social influence and analyze the correlation between influence and social actions.
My research group has developed an academic search system named Arnetminer (http://arnetminer.org), in which we aim to provide comprehensive services for academic communities. Currently, the system contains information of over 1 million researchers, 3 million academic papers (including 1 million full papers), and 8 thousand conferences. Services such as expertise search, citation tracing analysis, topical graph search, and topic browser are provided in Arnetminer. The system is in operation on the internet for four years and has received a large amount of accesses from 190 countries. We have applied for 7 patents based on the technologies developed in this system.
Prof. Bin Xu
Department of Computer Science and Technology inTsinghua University
Homepage:http://keg.cs.tsinghua.edu.cn/persons/xubin/
I received the BS, MS and DE degrees in Tsinghua University in 1996, 1998, and 2006. My research group has addressed fundamental problems of service computing, designed efficient algorithms for automatic service composition, and developed various service composition systems. As service-oriented computing (SOC) becomes the new computing paradigm of the Internet, it is a big necessity to automatically combine some web services in a certain way to satisfy specific requirements. My research achieved remarkable improvement on efficiency of automatic service composition.
My research group observes the data correlations between services' outputs and inputs. In existing works, data correlations are usually expressed in an implicit and automatic way by reasoning on data models or ontology models used in service definitions. My group proposes to express data correlations by explicit declarations. Referring to the fundamental idea of hyperlinks, the Service Data Link model (SDL) is proposed to explicitly declare data correlations among services. Furthermore, my group combines the expressive power of the SDL model with the Service Dependency Graph model (SDG), and present SDG+, an extended version of SDG, which introduces three extensions: 1) attribute quantifier, 2) attribute transform, and 3) explicit dependency. Taking advantage of explicit dependency, a major extension of SDG+, my group developed a composition algorithm for IEEE CEC Web Service Challenge 2007, which significantly improved the performance in dealing with datasets that contain complex inheritances among message types. The system designed based on this algorithm won the championship of Performance Evaluation in the competition.
In our research of automatic composition problem of QoS-aware and non-predefined workflow, a theoretical data-driven and QoS-optimized service composition model is proposed. In order to acquire composition solutions with optimal end-to-end QoS, my group utilize the semantic information among data types to realize semantic matching in service composition, and have proposed two composition algorithms. One is based on iterative generation, which iteratively searches services and updates the optimal QoS values; the other is based on greedy procedure which always selects the services with optimal accumulated QoS values. The experiment results show that the two algorithms are able to solve the problem correctly and satisfy the efficiency requirement of runtime systems. The system designed based on these two algorithms won the first place team and first runner-up team of Performance Evaluation in IEEE CEC Web Service Challenge 2008 and 2009.
Prof. Min Zhang
Department of Computer Science and Technology In Tsinghua University
Homepage:http://www.thuir.org/group/~mzhang
z-m@tsinghua.edu.cn
I received the BS, DE degrees in Tsinghua University in 1999 and 2003. My research interests are Web information retrieval (IR) and user behavior analysis. In most IR tasks, user queries are always short and fuzzy, while the known document space is huge with complex information. Therefore, the main problem in IR lies in the mismatch of information representations between the user query space and the known document space. This is what my work is abased on. My research contributions are: In the research of IR approaches to better document modeling, we studied the nature of novelty information, focusing on how to find new and non-redundant information to a user's request. We have proposed a document refinement approach based on query expansion, and a selective pooling-based document matching strategy. For document representation, an object-based document description is proposed. We also propose a unified generation model for topic relevance and opinion sentiments. The work has been published on important international journals and conferences, such as Journal of IR, SIGIR, CIKM, etc. One patent has been granted.
We are also leveraging wisdom of crowds and studying user behaviors based on large-scale user log data. Our proposed web page quality estimation and anti-spam algorithms are able to solve the robustness and timeliness problem of state-of-art technologies. Our automatic search engine performance evaluation approach handles the problems of the time and labor cost by human annotation. Currently, the automatic evaluation service for six main Chinese commercial search engines are daily provided online. We also study to understand user intentions, build user browsing graphs, and predict users' satisfaction level. Furthermore,we analyze the reliability of user behavior, and propose a model of user reliability and click relevance, which estimates the quality using both hot queries/clicks and long tail queries/clicks. This study has been published on important international journals and conferences, such as JASIST, WWW, WSDM, CIKM, etc. 8 patents have been filed, four of which were granted. Our proposed approaches have also been deployed on Sogou online search engine via "Tsinghua-Sohu" Joint Research Lab, and demonstrated good results and high potential in real web applications.
Several demos of their research projects can be found at Search E Search Engine Performance Evaluation System (http://searche.thuir.cn/), Top News Online Service (http://news.thuir.org), and Sogou Laboratory Homepage (www.sogou.com/labs).
Prof.Jia Jia
Department of Computer Science and Technology in Tsinghua University
jjia@mail.tsinghua.edu.cn
Jia JIA is an assistant professor at the Department of Computer Science and Technology, Tsinghua University. She got bachelor degree at Tsinghua University in 2003, and received the Ph.D. degree from Tsinghua University in 2008. She serves as the secretariat member of Professional Committee of Speech in Chinese Information Processing Society, and also a committee member of Multimedia Federation in China Society of Image and Graphic. She is the principal investigator of the National Basic Research (973) Program, National High-tech R&D (863) Program, NSFC project, and has a wide collaboration with Siemens, Dolby, IBM, Huawei and Tencent, etc. She has been awarded Scientific Progress Prizes from the National Ministry of Education in 2009. Her main research interests include Chinese speech processing, affective computing, and computational speech perception. She has authored more than 30 articles in leading international journals and conferences including IEEE Transaction on audio, speech and language processing, Neurocomputing, ICIP, ICB and workshop of ICCV.
Prof. Minlie Huang
Department of Computer Science and Technology
Email:aihuang@tsinghua.edu.cn
I received Bachelor of Engineering Physics in 2000 and Ph.D. in Computer Science in 2006 inTsinghua University . My research interest includes natural language processing, information diversity and summarization, opinion and review mining, question answering, and bioinformatics. My publications appear in top-ranking conferences such as ACL, AAAI, IEEE ICDM, NAACL, and PAKDD, and journals such as Genome Biology, Bioinformatics, BMC Bioinformatics, and International Journal of Biomedical Informatics.
During the past years, I and my group have been participating in TREC/TAC Competitions (organized by NIST of USA), which are standard evaluation conferences on information retrieval and information processing. In particular, we have achieved leading results on tasks of document summarization, text entailment, and knowledge linking. We have proposed a new approach to summarizing multiple documents, which is based on information distance theory. We also proposed novel methods and algorithms for opinion ranking, opinion classification, and multi-label classification. As of bioinformatics applications, we've made progresses on recognizing molecular entities (such as gene, protein, and disease), extracting semantic relations (gene-disease and protein-protein interactions), and modeling biomedical text.
I was a visiting fellow in University of Hamburg, Germany in 2006. In 2010, I was invited to visit and conduct research at the National Center for Biotechnology Information (NCBI) in the United States from January to July.
Prof. Jianyong Wang
Department of Computer Science and Technology in Tsinghua University
Homepage:http://dbgroup.cs.tsinghua.edu
I received Bachelor of Computer Science in in 1991 in Lanzhou University; And I got Master of Computer Science in Beijing University of Technology in1996; Getting Ph.D. in Computer Science in Chinese Academy of Sciences in1999.My research group focuses on various data mining algorithms and their novel applications, with particular interests in graph data mining, sequence data mining, Web data management and mining, uncertain data mining, and stream data mining.
In graph data mining, we investigate the problems such as coherent sub-graph mining, graph generator mining for classification, and structural anonymization of graph data (joint work with IBM). We have proposed the CLAN algorithm for mining closed cliques, the COCAIN and COCAIN* algorithms for mining closed coherent sub-graphs from large dense graph databases, and the FOGGER algorithm for graph generator discovery, which has shown high utility in classification (e.g., image classification and chemical compound classification).
We mainly study the problems in sequence data mining topic such as closed sequential pattern mining, gap-constrained sequential pattern mining, sequence generator pattern mining, summarization subsequence mining for clustering, and sequential pattern-based XML document clustering (joint work with IBM). We have proposed a series of efficient algorithms for these problems, including BIDE+, Gap-BIDE, FEAT, XProj, and CONTOUR. Theyalso explore the applications of these algorithms to areas such as classifying consumer product reviews, erroneous sentence detection, and protein data classification.
Web data management and mining attracts much attention of his group. In this topic, we mainly focus on the problems such as name disambiguation in digital library and Web people search, community detection in large networks, Web video topic detection, tracking and recommendation, entity and relationship extraction, large-scale dynamic index maintenance, and personalized recommendation systems. We have proposed two graph-based frameworks, GHOST and GRAPE, for name disambiguation in digital library and Web people search respectively, and a parallel community detection algorithm on large networks.
My group also works on problems in areas such as stream data mining and uncertain data mining. We have extended several typical frequent itemset mining algorithms for uncertain data (joint work with IBM), and proposed an efficient algorithm to directly mine discriminative patterns for classifying uncertain data and an algorithm for efficient item-set generator discovery over a stream sliding window.
Prof. Yongjin Liu
Department of Computer Science and Technology
Homepage:http://cg.cs.tsinghua.edu.
I received Bachelor of Mechanoelectronic in Tianjin University in 1998; I got M.E.& Ph.D.degree in Hong Kong University of Science & Technology in1999 & 2003.
In my graduate studies and postdoc work between 1998 and 2004, I focused on computational geometry and developing efficient algorithms for practical industrial applications. I have proposed an optimal polynomial algorithm that requires a minimal number of setups of workpiece in 4-axis NC machining. The algorithm has been proved to offer an approximate solution with a tightly-bounded performance ratio to an NP-complete problem. I have also proposed an efficient polynomial algorithm that is general and able to answer a series of inquiries concerning the intersection relations between a movable convex object and a set of planar polygons, given the previously best results needing to invoke several independent algorithms with different configuration-space models. The robot industry has shown great interests in his general model, especially its application in motion planning. He also worked on the industrial application of reverse engineering, in which I designed an optimal O(nlogn) algorithm that reconstructs water-tight and feature-preserving triangular meshes from non-uniform and unorganized point data. I contributed to an out-of-core algorithm that can deal with huge meshes and guarantee the 2-manifoldness of mesh models. The output 3D mesh models of this algorithm are readily to be manufactured using techniques such as rapid prototyping.
Since 2004, I have been focusing on intelligent processing of visual media, synthesizing techniques in computer graphics, human-computer interaction, pattern recognition and artificial intelligence, and building practical industrial-strength systems. I have successfully developed a plush toy design system using editable sketching curves, which has inspiring results. Sophisticated toy models can be easily designed with this system by novices mimicking traditional paper-and-pencil-based 2D/3D sketches. This system has been commercialized by a Chinese company, bringing 100k USD's revenue every year. To address the issues of knowledge representation and reasoning across domains in concurrent engineering, I have also proposed an integration framework with a semantic feature model, which provides a practical solution to knowledge capture and can be consistently reused in different domains of concurrent engineering. Currently, I am building an ontology representation for visual media, in which methods of logical inference in artificial intelligence can be utilized to achieve efficient knowledge representation and data mining.
Prof. Lifeng Sun
Department of Computer Science and Technology
Homepage:http://media.cs.tsinghua.edu.cn/en/sunlf
I got Bachelor &Ph.D.degree in National University of Defense Technology in1995&2000.
My research group addresses fundamental problems of 3D video representation, interaction, codec and transmission, carrying the objective of providing interactive immersive multimedia application in future Internet.
We built a statistical link model for application- layer multicast and discovered that the congestion of the neighboring link layer has Markov properties. By studying the logs of practical VoD broadcasting systems and P2P live broadcasting systems, my group established a model for users' online time and their prospective activities during that time. We also theoretically analyzed the bandwidth utilization optimality of pull-based p2p streaming systems, and proposed a push-pull model for p2p streaming data scheduling, which reduced the delay and control overhead, while maintaining optimal bandwidth utilization.
We proposed a multi-view video coding scheme, which supports random accesses among views. Compared with traditional methods, our scheme improves the coding efficiency up to 2-3 db, while reducing the delay of view switching. Based on the theory of directed acyclic graph, my group designed a spatial-temporal data paralleling algorithm for multi-view video coding over multi-core processor architecture, which achieves an approximate linear speed-up. Also, an accurate estimation model for the multi-core media processing task scheduling is established by abstracting the critical properties of multimedia applications and multi-core processors. Utilizing the correlation of data between different views, I proposed a collaborative hybrid overlay network structure for large-scale multi-view video streaming of 3DTV systems.
|
Contact us |
|
Jie Tang |