TAAI 2016 Keynote Talk: Intercultural Collaboration as a Multi‐Agent System
2 likes499 views
The document discusses multi-agent systems and their real-world applications, including security games, kidney exchange matching algorithms, stock market simulations, and tsunami evacuation decision making. It then examines intercultural collaboration as a new research field, providing examples of online translation tools and an agricultural support project in Vietnam. Finally, it analyzes collaboration protocols and collaborative translation in multi-agent systems for intercultural projects.
![Example2: Kidney Exchange
• First Idea 1986 [Rapaport]
• Exchange Started in 2003 [Roth, Sönmez, Ünver, …]
• United Network for Organ Sharing (UNOS) adopted
Sandholm’s algorithm in 2008 and executes it twice a
week.
Tuomas Sandholm
Carnegie Mellon UniversityIn the US, more than 70,000 patients are waiting
for kidney transplants, but only 10,000
transplants are carried out each year.
30‐chain! [New York Times 2/18/2012]
Wife
Husband Mother
Daughter
Contribution to Medical Care](https://image.slidesharecdn.com/taai20161126webup-161128024110/85/TAAI-2016-Keynote-Talk-Intercultural-Collaboration-as-a-Multi-Agent-System-8-320.jpg)
TAAI 2016 Keynote Talk: Intercultural Collaboration as a Multi‐Agent System
- 3. Multi‐Agent Systems • The research area is often called Autonomous Agent and Multi‐ Agent System. • It models a human society as a collection of independent decision making agents. • It analyzes and visualizes behaviors of agents via Multi‐Agent Based Simulation.
- 4. Multiagent System: A Major Research Area of AI • IJCAI 2016 （ Int. Joint Conf. on Artificial Intelligence） • Machine Learning: 22 sessions • Agent‐based and Multi‐agent Systems: 12 • Knowledge Representation, Reasoning, and Logic: 10 • Planning and Scheduling: 10 • Textbook of Artificial Intelligence Overview of the book The main unifying theme is the idea of an intelligent agent. We define AI as the study of agents that receive percepts from the environment and perform actions……
- 5. DAIWS 1980 Launched by Researchers in the United States ICMAS 1995 International Conference integrating workshops in the United States, Europe, and Japan PRIMA 1998 Asia Pacific International Conference AAMAS 2002 A+ Ranked International Conference MACC（JP） MAAMAW（EU） DAIWS（US） ICMASPRIMA (Workshop) AAMAS AA（US） ATAL（EU） PRIMA (Conference) Asia International 1980 2000 1990 2010 How We Developed Our Research Community
- 6. Multi‐Agent Systems are Getting Real Theory（Research for Computational Models of Human Society ） PRACTICE（Solving Problems for Society） １９８０ １９９０ ２０００ ２０１０ ２０２０ Multi‐Agent Based Simulation （Economy, Disaster Management） Matching Algorithm （Medical Care） Collaboration Distributed Search Self‐Organization Multi‐Agent Based Simulation High Speed Computing Big Data Negotiation Market Auction Game Theory Matching Algorithm Game Theory （Security）
- 7. Target #1 Target #2 Target #1 4, ‐3 ‐1, 1 Target #2 ‐5, 5 2, ‐1 Example1: Security Game Milind Tambe University of Southern California How to allocate limited security resources • Defenders commit first, and then Attackers observe defenders’ behavior and make decisions. • Optimal allocation: Weighted random Stackelberg Equilibrium Contribution to Security 29/52 23/52 22/52 30/52 Utility=0.433 Utility=0.294 Deployed at Los Angeles International Airport Attacker Defender 1000 Flights, 20 air marshals 10 41 combinations
- 8. Example2: Kidney Exchange • First Idea 1986 [Rapaport] • Exchange Started in 2003 [Roth, Sönmez, Ünver, …] • United Network for Organ Sharing (UNOS) adopted Sandholm’s algorithm in 2008 and executes it twice a week. Tuomas Sandholm Carnegie Mellon UniversityIn the US, more than 70,000 patients are waiting for kidney transplants, but only 10,000 transplants are carried out each year. 30‐chain! [New York Times 2/18/2012] Wife Husband Mother Daughter Contribution to Medical Care
- 9. Price Share of Tokyo Stock Exchange Tick Size • If Tick Size (minimum unit of ordering price) is too large, stock prices become unstable. • A newly discovered threshold made stock prices stable. • From July 2014, Tokyo Stock Exchange has changed tick sizes of TOPIX100 stocks. This impacted 80% of stock exchanges in Japan. Example 3: Designing Market Systems by Multi‐Agent Based Simulations Kiyoshi IZUMI University of Tokyo Contribution to Economy 1000 agents with different strategies
- 10. Example 4: Decision‐Making for Tsunami Evacuation by Multi‐Agent Based Simulations • Background • 5,781 people are living with risk of tsunami by earthquake. • Maintain escape routes to places over 2 meters above sea level. • Result • Determine bridges for reinforcement. • Determine the priority of streets to remove snow. 10 Itsuki Noda National Institute of Advanced Industrial Science and Technology Contribution to Disaster Management ① ② ③ ④⑤ ⑦ ⑥ ⑧ ⑨ ⑩ ⑪ Select bridges for reinforcement from 11 bridges Condition 1： Prioritize 11 streets to remove snow → 211 × 211 ＝ 4,194,304 cases Condition 2：
- 12. Intercultural Collaboration • After 9.11 we created a concept of intercultural collaboration, where participants in the same problem field with different cultures and languages work together towards the shared goal. • Because the research target emphasizes collaboration rather than communication, we can clearly identify research objectives; goal‐ directed group activities can be evaluated both qualitatively and quantitatively. 2001
- 13. To develop open source software by using machine translation. China Team Korea Team ICE2002 Organizers Translation Checker Japan Team Malaysia Team Intercultural Collaboration Experiment (ICE 2002)
- 14. Translation among Japanese, Chinese, Korean, Malay, and English 31,000 messages in one year. Message Input Translation Button Translated Massage Chinese Japanese Korean Malay English Multi‐Language BBS 2002
- 15. Translation Pentagon for ICE2002 Arcnet/sangenjaya Japanese English Malay Chinese Korean J ‐server Hard to collect translation engines to cover five languages. Hard to understand their contracts. Hard to evaluate their services. Hard to customize provided services. 2002
- 16. The Language Grid Fragmentation and recombination is the key to create customized multi‐language environments for various intercultural activities. more more Disaster Management Education Medical Care Sharing Multilingual Information Universal Playground Multilingual Reception Language Support for Multi-language and Multi-cultural Communities Sharing language resources such as dictionaries and machine translators around the world German Research Center for Artificial Intelligence Stuttgart University National Research Council, Italy Chinese Academy of Sciences NTT Research Labs Asian Disaster Reduction Center Kookmin University Princeton University NECTEC University of Indonesia Google. Inc.Kyoto University NICT 2007
- 17. Dictionary (Data) Parallel Text (Data) Human Interpreter Machine Translation Wrapping Wrapping Wrapping Wrapping Dictionary Dictionary Service Parallel Text Parallel Text Service Machine Translator Machine Translation Service Human Translator Human Translation Service Accurate word translation Approximate translation Low quality high speed translation High quality low speed translation 避難場所 disaster shelter 避難場所は、家から近い学校です。 In the case of disaster, people should be evacuated to a school nearby their house. 避難場所は、家から近い学校です。 Disaster shelter is school close from a house. 避難場所は、家から近い学校です。 Your disaster shelter is the school closest to your house. From Language Resources to Language Services
- 18. Language Grid Operation Centers Kyoto Operation Center (Kyoto Univ.，2007 ‐ ) Jakarta Operation Center (Univ. of Indonesia, 2012 ‐ ) Xinjiang Operation Center (Xinjiang Univ., 2014‐ ) East Asian Languages South‐East/South Asian Languages Central Asian Languages Share 225 language services from 170 groups in 22 countries Bangkok Operation Center （NECTEC, 2010 ‐ ） NSF LAAPS Grid Project LDC EU Meta Share Project ELRA/ELDA Federated Operation of Language Grid Servers
- 20. Overview of the Project • Organizations – Kyoto University: Multilingual Communication – NPO Pangaea: Education, Activities – University of Tokyo, Mie University : Agricultural Knowledge – Vietnam National University：Local Arrangements – MARD, DARD: Planning and Controlling Experiments • Schedule – 2011/02 ~ 2011/03 1st Experiment (Thien My, Vinlong Province) – 2012/10 ~ 2013/01 2nd Experiment (Thien My, Vinlong Province) – 2013/09 ~ 2014/01 3rd Experiment (Thien My and Dong Thanh, Vinlong Province) – 2014/02 ~ 2011/06 4th Experiment (Dong Thanh, Vinlong Province) Observation Recording Question/Answer
- 21. Example of Experiment Schedule 2012/09 2012/10 2012/11 2012/12 2013/01 Participants: 15 Youth Sowing Rice planting Rice reaping Rice Growing YMC Exp. Preparing Experiment Workshop 2012/10/29 2012/12/15 2012/12/12012/10/19
- 22. Designing Language Communication Japanese: たんぼの準備として、田起こしや代掻き、あぜぬりをして下さ い。 Vietnamese: Chuẩn bị đất là kết hợp giữa canh tác đất, cày bừa và đắp bờ. (English: Land preparation is a combination of tillage of the soil, puddling and levee painting.) Example-based Translation Service Translation Service with Agricultural Dictionary Japanese- English Translator English- Vietnamese Translator Multilingual Dictionary for Agriculture ☆ Japanese Part-of-Speech Tagger Vietnamese Part-of-Speech Tagger Multilingual Parallel Texts for Agriculture ☆☆ Example- Based Machine Translator Japanese Dependency Parser Back Translation Best Translation Selection Online Consultation by Human Vietnamese Farmers Japanese Experts ☆ Multilingual Dictionary for Agriculture is provided by NPO Pangaea, Japan National Agriculture Research Center, Vietnam MARD. Entry Number: 3,099 (Sep. 2014) ☆☆ Multilingual Parallel Texts for Agriculture is provided by NPO Pangaea, Japan National Agriculture Research Center, Vietnam MARD. Entry Number: 2,485 (Sep. 2014) Best Translation Selection Service
- 25. Self‐initiated repair or user adaptation to machine translation: Before posting a message, the sender modifies input texts to improve translation quality. Other‐initiated repair or collaborative translation: If a translation was imperfect, collaboration is initiated by a receiver. Topic 1: Collaboration Protocol Interaction patterns among users in ICE2002. S. Nomura, T. Ishida, N. Yamashita, M. Yasuoka and K. Funakoshi. Open source software development with your mother language: Intercultural Collaboration Experiment 2002. International Conference on Human‐Computer Interaction, Vol. 4, pp. 1163‐1167, 2003. 2002
- 26. Collaborative Translation Protocol Machine Translator Source Language Side Target Language Side 1. Send the source sentence 2. Evaluate fluency of the translation. 3. Send the sentence which is modified to be fluent. 4. Evaluate adequacy of the translation against the source sentence. Improve adequacy of the translation result Improve fluency of the translation result D. Morita and T. Ishida. Collaborative translation by monolinguals with machine translators. International Conference on Intelligent User Interfaces, pp. 361‐366, 2009. 2009
- 29. Topic 3: Problem Solving Organization Japanese Experts Vietnamese Youths Vietnamese Farmers NPO Pangaea Univ. of Tokyo Kyoto Univ. MARD Mie Univ. Knowledge Communication Organizational Communication Japanese Bridger English Bridger Language Communication JAVI EN Vietnamese Bridger VNU DARD Agricultural Data Agricultural Knowledge Machine Translation 2015
- 30. Multi‐Agent Based Participatory Simulation Analysis Design Field Experiment or Multi-Agent Based Participatory Simulation Language / Knowledge Communication ModelingMachine Translatio n (JA→EN) Post- Editing (EN→EN) Japanese Bridger Vietnamese Bridger Vietnamese Youth Japanese Expert Machine Translation (EN→VI) Parallel Texts （JA→VI） Pre- Editing (JA→JA ) English Bridger Post- Editing (VI→VI) 2015 Test technologies in advance with various stakeholders.
- 33. Thank You! Special Thanks to: Tuomas Sandholm,Milind Tambe, Bo An, Itsuki Noda, Kiyoshi Izumi, Shigeo Matsubara, Yohei Mukakami, Donghui Lin, Yumoko Mori, Toshiyuki Takasaki