WI 2007 Notes of Jie Bao

From Tetherless World Wiki

Jump to: navigation, search

To view this tutorial, you need to use Firfox and install Greasemonkey, and Wikipedia Presentation script. Don't be panic, those can be easily done by several mouse clicks. When you see "Start presentation" under the title, you are all set!

International Conference on Web Intelligence 2007 (Program), by Jie Bao



Contents

Friday, November 2, 2007

  • Tutorial on Agent Mediated Knowledge Management (abstract)
    Dr. Virginia Dignum, Utrecht Universit, The Netherlands
    • A high-level talk, no much technical details.
    • Argues for distributed knowledge management (KM), since it is more natural (against centralized KM).
      • Respect distributed nature of knowledge
      • Inherent goal dichotomy between business processes and KM
      • KM is "wicked problem solving" (social process, no a priori solution, interaction needed)
      • has to deal with changing environments (Agile structure)
    • AMKM wokshops (03-05): [1], 04, 05
  • Tutorial: Distributed Constraint Reasoning - A Paradigm for Effective Coordination in Multiagent Systems (abstract)
    Makoto Yokoo, Kyushu University, Japan
    Jörg Denzinger, University of Calgary, Canada
    Marius Silaghi, Florida Tech, USA
    Adrian Petcu, Swiss Federal Inst. of Tech.
    • Constraint Reasoning, e.g. 8-queen problem
      • Constraint Satisfaction, constraint optimization, constraint reasoning
    • Distributed Constraint Reasoning
      • Agents has intra- and inter-agent constraints
      • (Jie: modular ontology reasoning is closed to distributed constraint satisfaction)
    • Semi-cooperative agents - privacy, individual interests
      • (Jie: more interesting, close to the distributed privacy-persevering reasoning problem)
    • Be careful! Communication and cooperation overhead vs. Efficiency gain from parallel processing
      • (Jie: discuss this in our P-DL reasoning paper (e.g. Message cache). That's why we do not provide a particular communication protocol for such a reasoner so far: it should determined based on experimental results.)
    • Example approach: Asynchronous Backtracking [YokDurIshKuw92]
      • Avoid infinite processing loops (Jie: We have the problem in P-DL!)
    • Example approach: Weak-commitment Search [Yokoo 94]
    • Backtrack Search (exp. # of messages) vs. Dynamic Programming (linear # of messages)
      • Example: ADOPT (search) vs. DPOP (DP) [the comparison table is neat]
    • Privacy is the main reason for distributed constraint reasoning!
    • Ideas:
      • (Jie: Distributed reasoning with DL under the conventional semantics may be possible if each ontology is "local" '(Grau's notion))
      • (Jie: The current P-DL reasoning approach is based on backtracking. However, I believe dynamic programming aproaches may also be explored, i.e. Storing results of the solved "smaller" problems. It is close to the idea of "look up table" for messages we have. )
      • (Jie: Incomplete DPOP algorithms looks interesting. Is it also useful in DL reasoning? At least for ABox instance retrieval
  • Analysis of Protein-Protein Diverse Interfaces
    Feihong Wu, Iowa State University (go! Cyclone) (this is a BIBM 2007 paper)
    • Problem: Identifying PPI from protein sequence
    • Solution:
      • Interface propensity analysis (what is IP? );
      • Features used: Side Chain Orientation (<0.5pi),Surface Roughness(<22), Solid Angle(1.8pi,2.2pi), Cx Value (Protrusion), Surface Microenvironment - Hydrophobicity (-4.5,-2.5), Interface Size (>2 neighbours are also PPI): compute IP from them.
    • Highpoint: Find some features of PPI; Large dataset (from PPIDB).
    • Questions:
      • Homo vs hetero?
      • Chemcial or Biological interface? (this paper deal with chemical property)
  • We skipped most of the afternoon session to see Flavian. He is tough and recovers pretty well from the accident.
  • I have two lunches today. (I had typed the story but lost the whole text by unsaving. Give up.)

People met: Chunming Chen, University of South Carolina; Shenghui Wang, Vrije Universiteit Amsterdam

Saturday, November 3, 2007

Morning

  • In search of the Golden Rules of Human-Automation Teamwork
    Jeffrey M. Bradshaw, IHMC
    • It is a replacement of "How Relevant is Game Theory to Intelligent Agent Technology?" by Yoav Shoham
    • Nice robot video
  • WI/IAT Joint Keynote: Computer Science in the 21st Century
    Dieter Fensel, University of Innsbruck/Digital Enterprise Research Institute, Austria
    • Concept: Service Ware, SOA (Service-oriented Architecture)
      • there are 12k web services now, see seekda.com
      • but finally there will be billions of services, thus here comes the problems of openness, heterogeneity, distributedness, scalability
      • Reasoning on the web; LarKc, Ning Zhong
      • Service discovery - remember that we will have billions of them, how to rank by utility?
    • Semantic Web
      • data: RDF, SW cake (usual introduction)
      • processes: SAWSDL (semantic annotation of WSDL), WSMO, OWL-S, SWSF (Semantic Web Services Framework)
    • Boundaries of Semantics
      • large heterogeneous, distributed systems
      • Self representation and self reflection
      • classical ACID of database is no longer true (incomplete, inconsistent)
      • Heuristic approach - give up completeness and consistency, for the sake of scalability; Approximate solutions
    • Q & A:
      • question: "why SW had hard time to take off?" (a sharp question!) Answer: (Jie- I don't understand )
      • question: where is science, in addition to engineering issues (e.g., scalability tricks). answer: for example, the logics

Afternoon

  • Industry Keynote: Enterprise Information Mashups: Integrating Information, Simply
    Anant Jhingran, VP and CTO, IBM Silicon Valley Laboratory
    • (leave in middle for another session)
  • Analysis of multi-actor policy contexts using perception graphs
    Pieter W.G. Bots, Delft University of Technology, France
    • (I thought it is a "policy" paper in the web sense. However, policy there means goal-oriented decision making.)
    • Problem: "gain better understanding of the behavior of actors in policy making processes"
    • Solution:
      • Dynamic Actor Network Analysis (DANA)
      • Perception graph: like casual network
  • Industry Keynote: Dataspaces -- Enabling the Next Generation Data Management Applications
    Alon Halevy, Research Scientist, Google
    • Problem: New data management needed on Web 2.0
      • Some examples of the data integration problem (RDB, Web)
      • Deep web: data that is only accessible by HTML forms
        • Deep web is impractical for web search, neither the traditional data integration (mediated schema is almost impossible)
      • Typically, data integration is a process with huge $$$$ investment, but no benefit for a long period
    • Solution: dataspace
      • Get immediate gain on data integration.
      • Dataspace enhancement: automatic detect relationships [CIDR05]; reference reconciliation [SIGMID 05]; combining structured and unstructured querying [WebDB 06]; Visualization
      • Current Solution: Crawl deep web [Madhavan, Kot, Rasmussen]
        • put result pages in the index; let the ranking infrastructure take care of the rest.
      • Google's Dataspaces Technology Dings PageRank by Clint Boulton, eWeek, October 31, 2007.
      • Google base (e.g. recipe search): simple schema-based search on google
  • Industry Keynote: Social and Semantic Structures in Web Search
    Andrew Tomkins, Chief Scientist of Search, Yahoo

Others

  • Trivia: Second day of no internet in any meeting room

Sunday, November 4, 2007

Morning

  • A Unified Approach to Researcher Profiling
    Limin Yao, Jie Tang, and Juanzi Li - Tsinghua University
    • Problem: build a semantic profile for a researcher from web resource (e.g., homepage)
    • Solution:
      • 3 steps: strelevant page finding (e.g., by Google), preprocessing (segment text intotokens and assign possible tags to each token), and tagging.
      • Schema extended from FOAF
      • employ Conditional Random Fields (CRF) (a probability model) as the tagging model.
    • Links:
      • ArnetMiner
      • KEG_CRF
      • DBLife ( Developed by Database Group at University of Wisconsin and Community Systems Group at Yahoo! Researc)
    • Remark: impressive, that's one thing I'm thirsty for to extend this wiki.


Afternoon

  • Document-Centric Query Answering for the Semantic Web
    Yuanbo Guo, Jeff Heflin -Lehigh University
    • Problem: can we query a subset of a large knowledge base (that contains a set of document)?
      • each document is an ABox
    • Solution:
      • Types of queries:
        • Boolean document entailment (BDE), e.g., is x of C?
        • Boolean document provenance (BDP), e.g., what subset of the document supports A(x)
        • Retrieval document entailment (RDE), e.g., return the instance set of C?
        • Retrieval document provenance (RDP): retrieves all the individuals that satisfy C (query condition) and in addition the minimal consistent subsets of D that support each answer.
      • Key point: Preprocess (ABox summarization - to a simpler ABox) and Partition document
    • Remark: (Jie) the partitioning of ABox part looks similar to ABox modularization. There is a paper this year on DL workhop.
  • Local and On-the-fly Choreography-based Web Service Composition
    Saayan Mitra, Samik Basu, Ratnesh Kumar - Iowa State University
    • Problem: Given a set of services and a goal, whether a choreographer exists which can act as an intermediary to realize the goal from the set of services, and if exists, how to find it.
    • Solution: services and goal are modeled as i/o automata; the solution mechanism is to find all possible choreographed behaviors of services by interleaving product and taking the transduced closure to form the Universal Automaton, then checking whether the goal is simulated by the Universal Automaton; implementation of the technique proceeds in a goal-directed fashion generating the Universal Automaton on-the-fly.
  • Towards a Media Interpretation Framework for the Semantic Web
    S. Espinosa Peraldi, A. Kaya, S. Melzer, R. Moller, M. Wessel - Hamburg University of Technology, Germany
    • Problem: maximize precision and recall of semantics-based information retrieval (of image)
    • Solution: a framework that leverages low-level information extraction (e.g. pattern) to a higher level of abstraction; enables the automatic annotation of documents through high-level content descriptions; enable information retrieval using more abstract terms

Trivia


Monday, November 5, 2007


  • In the afternnon, we went to San Jose.
Facts about WI 2007 Notes of Jie BaoRDF feed
About International Conference on Web Intelligence 2007  +
Has author Jie Bao  +
Has year 2007  +
Retrieved from "http://tw.rpi.edu/wiki/WI_2007_Notes_of_Jie_Bao"
Personal tools