From nasmith at cs.cmu.edu Sat Feb 5 12:29:08 2011 From: nasmith at cs.cmu.edu (Noah A Smith) Date: Sat, 5 Feb 2011 12:29:08 -0500 Subject: [Intelligence Seminar] Intelligence Seminar: Eugene Charniak, 2/8 1:30pm GHC 4405, "Top-Down Nearly-Context-Sensitive Parsing" Message-ID: INTELLIGENCE SEMINAR FEBRUARY 8 AT 1:30pm, IN GHC 4405 (PLEASE NOTE THE UNUSUAL ROOM AND TIME) SPEAKER: EUGENE CHARNIAK (Brown University) Host: Noah Smith For meetings, contact Dana Houston (dhouston at cs.cmu.edu) TOP-DOWN NEARLY-CONTEXT-SENSITIVE PARSING We present a new syntactic parser that works left-to-right and top-down, thus maintaining a fully connected parse tree for a few alternative parse hypotheses. All of the commonly used statistical parsers use context-free dynamic programming algorithms and as such work bottom up on the entire sentence. Thus they only find a complete fully connected parse at the very end. In contrast, both subjective and experimental evidence shows that people understand a sentence word-to-word as they go along, or close to it. The constraint that the parser keeps one or more fully connected syntactic trees is intended to operationalize this cognitive fact. Our parser achieves a new best result for top-down generative parsers of 89.4%, a 20% error reduction over the previous result for parsers of this type of 86.8% (Roark, 2001). The improved performance is due to embracing the very large feature set available in exchange for giving up dynamic programming. BIO Eugene Charniak is University Professor of Computer Science and Cognitive Science at Brown University and past chair of the Department of Computer Science. He received his A.B. degree in Physics from University of Chicago, and a Ph.D. from M.I.T. in Computer Science. He has published four books, the most recent being Statistical Language Learning. He is a Fellow of the American Association of Artificial Intelligence and was previously a Councilor of the organization. His research has always been in the area of language understanding or technologies which relate to it. Over the last 20 years he has been interested in statistical techniques for many areas of language processing, including parsing and discourse. -------------- next part -------------- An HTML attachment was scrubbed... URL: http://mailman.srv.cs.cmu.edu/pipermail/intelligence-seminar-announce/attachments/20110205/ffe7d66f/attachment.html From nasmith at cs.cmu.edu Mon Feb 7 14:58:02 2011 From: nasmith at cs.cmu.edu (Noah A Smith) Date: Mon, 7 Feb 2011 14:58:02 -0500 Subject: [Intelligence Seminar] Intelligence Seminar: Eugene Charniak, TUESDAY 2/8 1:30pm GHC 4405, "Top-Down Nearly-Context-Sensitive Parsing" Message-ID: INTELLIGENCE SEMINAR FEBRUARY 8 AT 1:30pm, IN GHC 4405 (PLEASE NOTE THE UNUSUAL ROOM AND TIME) SPEAKER: EUGENE CHARNIAK (Brown University) Host: Noah Smith For meetings, contact Dana Houston (dhouston at cs.cmu.edu) TOP-DOWN NEARLY-CONTEXT-SENSITIVE PARSING We present a new syntactic parser that works left-to-right and top-down, thus maintaining a fully connected parse tree for a few alternative parse hypotheses. All of the commonly used statistical parsers use context-free dynamic programming algorithms and as such work bottom up on the entire sentence. Thus they only find a complete fully connected parse at the very end. In contrast, both subjective and experimental evidence shows that people understand a sentence word-to-word as they go along, or close to it. The constraint that the parser keeps one or more fully connected syntactic trees is intended to operationalize this cognitive fact. Our parser achieves a new best result for top-down generative parsers of 89.4%, a 20% error reduction over the previous result for parsers of this type of 86.8% (Roark, 2001). The improved performance is due to embracing the very large feature set available in exchange for giving up dynamic programming. BIO Eugene Charniak is University Professor of Computer Science and Cognitive Science at Brown University and past chair of the Department of Computer Science. He received his A.B. degree in Physics from University of Chicago, and a Ph.D. from M.I.T. in Computer Science. He has published four books, the most recent being Statistical Language Learning. He is a Fellow of the American Association of Artificial Intelligence and was previously a Councilor of the organization. His research has always been in the area of language understanding or technologies which relate to it. Over the last 20 years he has been interested in statistical techniques for many areas of language processing, including parsing and discourse. -------------- next part -------------- An HTML attachment was scrubbed... URL: http://mailman.srv.cs.cmu.edu/pipermail/intelligence-seminar-announce/attachments/20110207/e076c788/attachment.html From nasmith at cs.cmu.edu Mon Feb 7 15:17:41 2011 From: nasmith at cs.cmu.edu (Noah A Smith) Date: Mon, 7 Feb 2011 15:17:41 -0500 Subject: [Intelligence Seminar] Intelligence Seminar: Robin Hanson, 2/15 3:30pm GHC 4303, "The Potential of Prediction Markets" Message-ID: INTELLIGENCE SEMINAR FEBRUARY 15 AT 3:30PM, IN GHC 4303 SPEAKER: ROBIN HANSON (George Mason University) Host: Tuomas Sandholm For meetings, contact Charlotte Yano (yano at cs.cmu.edu) THE POTENTIAL OF PREDICTION MARKETS Prediction markets, and related forms of collective forecasting, are slowly becoming more robust and reliable, and applied more widely in industry. This talk reviews the concept, mechanisms, issues, and enormous future potential of related information aggregation mechanisms. BIO Robin Hanson is an associate professor of economics at George Mason University, a research associate at the Future of Humanity Institute of Oxford University, and chief scientist at Consensus Point. After receiving his Ph.D. in social science from the California Institute of Technology in 1997, Robin was a Robert Wood Johnson Foundation health policy scholar at the University of California at Berkeley. In 1984, Robin received a masters in physics and a masters in the philosophy of science from the University of Chicago, and afterward spent nine years researching artificial intelligence, Bayesian statistics, and hypertext publishing at Lockheed, NASA, and independently. Robin has pioneered prediction markets since 1988. He was the first to write in detail about people creating and subsidizing markets in order to gain better estimates on those topics. Robin was a principal architect of the first internal corporate markets, at Xanadu in 1990, of the first web markets, the Foresight Exchange since 1994, and of DARPA's Policy Analysis Market, from 2001 to 2003. Robin has developed new technologies for conditional, combinatorial, and intermediated trading, and has studied insider trading, manipulation, and other foul play. -------------- next part -------------- An HTML attachment was scrubbed... URL: http://mailman.srv.cs.cmu.edu/pipermail/intelligence-seminar-announce/attachments/20110207/a6d0a96f/attachment-0001.html From dhouston at cs.cmu.edu Fri Feb 11 16:13:07 2011 From: dhouston at cs.cmu.edu (Dana Houston) Date: Fri, 11 Feb 2011 16:13:07 -0500 Subject: [Intelligence Seminar] Feb. 15, 3:30pm: Presentation by Robin Hanson Message-ID: <4D55A663.2090003@cs.cmu.edu> INTELLIGENCE SEMINAR FEBRUARY 15 AT 3:30PM, IN GHC 4303 SPEAKER: ROBIN HANSON (George Mason University) Host: Tuomas Sandholm For meetings, contact Charlotte Yano (yano at cs.cmu.edu) THE POTENTIAL OF PREDICTION MARKETS Prediction markets, and related forms of collective forecasting, are slowly becoming more robust and reliable, and applied more widely in industry. This talk reviews the concept, mechanisms, issues, and enormous future potential of related information aggregation mechanisms. BIO Robin Hanson is an associate professor of economics at George Mason University, a research associate at the Future of Humanity Institute of Oxford University, and chief scientist at Consensus Point. After receiving his Ph.D. in social science from the California Institute of Technology in 1997, Robin was a Robert Wood Johnson Foundation health policy scholar at the University of California at Berkeley. In 1984, Robin received a masters in physics and a masters in the philosophy of science from the University of Chicago, and afterward spent nine years researching artificial intelligence, Bayesian statistics, and hypertext publishing at Lockheed, NASA, and independently. Robin has pioneered prediction markets since 1988. He was the first to write in detail about people creating and subsidizing markets in order to gain better estimates on those topics. Robin was a principal architect of the first internal corporate markets, at Xanadu in 1990, of the first web markets, the Foresight Exchange since 1994, and of DARPA's Policy Analysis Market, from 2001 to 2003. Robin has developed new technologies for conditional, combinatorial, and intermediated trading, and has studied insider trading, manipulation, and other foul play. -- Dana M. Houston Language Technologies Institute School of Computer Science Carnegie Mellon University 5407 Gates Hillman Complex 5000 Forbes Avenue Pittsburgh, PA 15213 T: (412)268-6591 F: (412)268-6298 From dhouston at cs.cmu.edu Tue Feb 15 10:09:49 2011 From: dhouston at cs.cmu.edu (Dana Houston) Date: Tue, 15 Feb 2011 10:09:49 -0500 Subject: [Intelligence Seminar] Feb. 15, 3:30pm: Presentation by Robin Hanson Message-ID: <4D5A973D.8090809@cs.cmu.edu> INTELLIGENCE SEMINAR FEBRUARY 15 AT 3:30PM, IN GHC 4303 SPEAKER: ROBIN HANSON (George Mason University) Host: Tuomas Sandholm For meetings, contact Charlotte Yano (yano at cs.cmu.edu) THE POTENTIAL OF PREDICTION MARKETS Prediction markets, and related forms of collective forecasting, are slowly becoming more robust and reliable, and applied more widely in industry. This talk reviews the concept, mechanisms, issues, and enormous future potential of related information aggregation mechanisms. BIO Robin Hanson is an associate professor of economics at George Mason University, a research associate at the Future of Humanity Institute of Oxford University, and chief scientist at Consensus Point. After receiving his Ph.D. in social science from the California Institute of Technology in 1997, Robin was a Robert Wood Johnson Foundation health policy scholar at the University of California at Berkeley. In 1984, Robin received a masters in physics and a masters in the philosophy of science from the University of Chicago, and afterward spent nine years researching artificial intelligence, Bayesian statistics, and hypertext publishing at Lockheed, NASA, and independently. Robin has pioneered prediction markets since 1988. He was the first to write in detail about people creating and subsidizing markets in order to gain better estimates on those topics. Robin was a principal architect of the first internal corporate markets, at Xanadu in 1990, of the first web markets, the Foresight Exchange since 1994, and of DARPA's Policy Analysis Market, from 2001 to 2003. Robin has developed new technologies for conditional, combinatorial, and intermediated trading, and has studied insider trading, manipulation, and other foul play. -- Dana M. Houston Language Technologies Institute School of Computer Science Carnegie Mellon University 5407 Gates Hillman Complex 5000 Forbes Avenue Pittsburgh, PA 15213 T: (412)268-6591 F: (412)268-6298 From dhouston at cs.cmu.edu Thu Feb 17 13:37:30 2011 From: dhouston at cs.cmu.edu (Dana Houston) Date: Thu, 17 Feb 2011 13:37:30 -0500 Subject: [Intelligence Seminar] Feb. 22, 3:30pm: Presentation by Bart Selman Message-ID: <4D5D6AEA.5060103@cs.cmu.edu> INTELLIGENCE SEMINAR FEBRUARY 22 AT 3:30PM, IN GHC 4303 SPEAKER: BART SELMAN (Cornell University) Host: Tuomas Sandholm For meetings, contact Charlotte Yano (yano at cs.cmu.edu) GOING BEYOND NP: NEW CHALLENGES IN INFERENCE TECHNOLOGY In recent years, we have seen tremendous progress in inference technologies. For example, in the area of Boolean satisfiability (SAT) and Mixed Integer Programming (MIP) solvers now enable us to tackle significant practical problem instances with up to a million variables and constraints. Key to this success is the ability to strike the right balance between the expressiveness of the underlying representation formalism and the efficiency of the solvers. The next challenge is to extend the reach of these solvers to more complex tasks that lie beyond NP. I will discuss our work on sampling, counting, probabilistic reasoning, and adversarial reasoning. In particular, I will discuss a new sampling technique based on the so-called flat histogram method, from statistical physics. The technique allows for fast probabilistic inference and learning in Markov Logic networks and other graphical models. In the area of adversarial reasoning, the UCT method, based on sampling strategies first developed for use in multi-armed bandit scenarios, provides a compelling alternative to traditional minimax search. The method has led to an exciting advance in the strength of GO programs. I will discuss insights into the surprising effectiveness of the UCT technique. BIO Bart Selman is a professor of computer science at Cornell University. His research interests include efficient reasoning procedures, planning, knowledge representation, and connections between computer science and statistical physics. He has (co-)authored over 150 papers, which have appeared in venues spanning Nature, Science, Proc. Natl. Acad. of Sci., and a variety of conferences and journals in AI and Computer Science. He has received six Best Paper Awards, and is an Alfred P. Sloan Research Fellowship recipient, a Fellow of AAAI, and a Fellow of AAAS. -- Dana M. Houston Language Technologies Institute School of Computer Science Carnegie Mellon University 5407 Gates Hillman Complex 5000 Forbes Avenue Pittsburgh, PA 15213 T: (412)268-6591 F: (412)268-6298 From eugenefink at cmu.edu Mon Feb 21 11:20:09 2011 From: eugenefink at cmu.edu (Eugene Fink) Date: Mon, 21 Feb 2011 11:20:09 -0500 (EST) Subject: [Intelligence Seminar] Feb. 22, 3:30pm: Presentation by Bart Selman Message-ID: INTELLIGENCE SEMINAR FEBRUARY 22 AT 3:30PM, IN GHC 4303 SPEAKER: BART SELMAN (Cornell University) Host: Tuomas Sandholm For meetings, contact Charlotte Yano (yano at cs.cmu.edu) GOING BEYOND NP: NEW CHALLENGES IN INFERENCE TECHNOLOGY In recent years, we have seen tremendous progress in inference technologies. For example, in the area of Boolean satisfiability (SAT) and Mixed Integer Programming (MIP) solvers now enable us to tackle significant practical problem instances with up to a million variables and constraints. Key to this success is the ability to strike the right balance between the expressiveness of the underlying representation formalism and the efficiency of the solvers. The next challenge is to extend the reach of these solvers to more complex tasks that lie beyond NP. I will discuss our work on sampling, counting, probabilistic reasoning, and adversarial reasoning. In particular, I will discuss a new sampling technique based on the so-called flat histogram method, from statistical physics. The technique allows for fast probabilistic inference and learning in Markov Logic networks and other graphical models. In the area of adversarial reasoning, the UCT method, based on sampling strategies first developed for use in multi-armed bandit scenarios, provides a compelling alternative to traditional minimax search. The method has led to an exciting advance in the strength of GO programs. I will discuss insights into the surprising effectiveness of the UCT technique. BIO Bart Selman is a professor of computer science at Cornell University. His research interests include efficient reasoning procedures, planning, knowledge representation, and connections between computer science and statistical physics. He has (co-)authored over 150 papers, which have appeared in venues spanning Nature, Science, Proc. Natl. Acad. of Sci., and a variety of conferences and journals in AI and Computer Science. He has received six Best Paper Awards, and is an Alfred P. Sloan Research Fellowship recipient, a Fellow of AAAI, and a Fellow of AAAS. From dhouston at cs.cmu.edu Wed Mar 9 12:07:47 2011 From: dhouston at cs.cmu.edu (Dana Houston) Date: Wed, 09 Mar 2011 12:07:47 -0500 Subject: [Intelligence Seminar] March 15, 3:30pm: Presentation by Itai Ashlagi Message-ID: <4D77B3E3.4030302@cs.cmu.edu> INTELLIGENCE SEMINAR MARCH 15 AT 3:30PM, IN GHC 4303 SPEAKER: ITAI ASHLAGI (Massachusetts Institute of Technology) Host: Tuomas Sandholm For meetings, contact Charlotte Yano (yano at cs.cmu.edu) INDIVIDUAL RATIONALITY AND PARTICIPATION IN LARGE SCALE, MULTI-HOSPITAL KIDNEY EXCHANGE As multi-hospital kidney exchange clearinghouses have grown, the set of players has grown from patients and surgeons to include hospitals. Hospitals have the option of enrolling only their hard-to-match patient-donor pairs, while conducting easily arranged exchanges internally. This behavior has already started to be observed. We show that the cost of making it individually rational for hospitals to participate fully is low in almost every large exchange pool (although the worst-case cost is very high), while the cost of failing to guarantee individually rational allocations could be large, in terms of lost transplants. We also identify an incentive-compatible mechanism. BIO Itai Ashlagi is an Assistant Professor at the Sloan School of Management. He is interested in mechanism design, market design, and game theory. In particular, he is interested in both developing and applying economic and optimization/CS tools for designing better marketplaces. Ashlagi is the recipient of the outstanding paper award at the ACM Conference on Electronic Commerce 2009. Before coming to MIT, he spent two years as a postdoctoral researcher at Harvard Business School. He was also a consultant researcher in 2010 for Microsoft Research in New England. He graduated from Technion-Israel Institute of Technology in 2008. -- Dana M. Houston Language Technologies Institute School of Computer Science Carnegie Mellon University 5407 Gates Hillman Complex 5000 Forbes Avenue Pittsburgh, PA 15213 T: (412)268-6591 F: (412)268-6298 From dhouston at cs.cmu.edu Mon Mar 14 10:55:53 2011 From: dhouston at cs.cmu.edu (Dana Houston) Date: Mon, 14 Mar 2011 10:55:53 -0400 Subject: [Intelligence Seminar] March 15, 3:30pm: Presentation by Itai Ashlagi - REMINDER In-Reply-To: <4D77B3E3.4030302@cs.cmu.edu> References: <4D77B3E3.4030302@cs.cmu.edu> Message-ID: <4D7E2C79.3030707@cs.cmu.edu> INTELLIGENCE SEMINAR > MARCH 15 AT 3:30PM, IN GHC 4303 > > SPEAKER: ITAI ASHLAGI (Massachusetts Institute of Technology) > Host: Tuomas Sandholm > For meetings, contact Charlotte Yano (yano at cs.cmu.edu) > > INDIVIDUAL RATIONALITY AND PARTICIPATION IN LARGE SCALE, > MULTI-HOSPITAL KIDNEY EXCHANGE > > As multi-hospital kidney exchange clearinghouses have grown, the set > of players has grown from patients and surgeons to include hospitals. > Hospitals have the option of enrolling only their hard-to-match > patient-donor pairs, while conducting easily arranged exchanges > internally. This behavior has already started to be observed. We show > that the cost of making it individually rational for hospitals to > participate fully is low in almost every large exchange pool (although > the worst-case cost is very high), while the cost of failing to > guarantee individually rational allocations could be large, in terms > of lost transplants. We also identify an incentive-compatible mechanism. > > BIO > > Itai Ashlagi is an Assistant Professor at the Sloan School of Management. > He is interested in mechanism design, market design, and game theory. In > particular, he is interested in both developing and applying economic and > optimization/CS tools for designing better marketplaces. Ashlagi is the > recipient of the outstanding paper award at the ACM Conference on > Electronic Commerce 2009. Before coming to MIT, he spent two years as a > postdoctoral researcher at Harvard Business School. He was also a > consultant researcher in 2010 for Microsoft Research in New England. He > graduated from Technion-Israel Institute of Technology in 2008. > > > -- Dana M. Houston Language Technologies Institute School of Computer Science Carnegie Mellon University 5407 Gates Hillman Complex 5000 Forbes Avenue Pittsburgh, PA 15213 T: (412)268-6591 F: (412)268-6298 From eugenefink at cmu.edu Mon Mar 14 17:18:36 2011 From: eugenefink at cmu.edu (Eugene Fink) Date: Mon, 14 Mar 2011 17:18:36 -0400 (EDT) Subject: [Intelligence Seminar] March 15 Presentation CANCELED Message-ID: Dear Intelligence Seminar Participants: WE WILL HAVE *NO* INTELLIGENCE SEMINAR TOMORROW (MARCH 15). With an apology for the short notice, the presentation of Itai Ashlagi tomorrow (March 15) has been CANCELED. We hope to reschedule his visit for a later day. Sincerely, Eugene Fink From dhouston at cs.cmu.edu Tue Mar 15 08:49:00 2011 From: dhouston at cs.cmu.edu (Dana Houston) Date: Tue, 15 Mar 2011 08:49:00 -0400 Subject: [Intelligence Seminar] CANCELED - March 15, 3:30pm: Presentation by Itai Ashlagi - CANCELED In-Reply-To: <4D7E2C79.3030707@cs.cmu.edu> References: <4D77B3E3.4030302@cs.cmu.edu> <4D7E2C79.3030707@cs.cmu.edu> Message-ID: <4D7F603C.5050908@cs.cmu.edu> The Intelligence Seminar for tomorrow has been canceled. On 3/14/2011 10:55 AM, Dana Houston wrote: > > INTELLIGENCE SEMINAR >> MARCH 15 AT 3:30PM, IN GHC 4303 >> >> SPEAKER: ITAI ASHLAGI (Massachusetts Institute of Technology) >> Host: Tuomas Sandholm >> For meetings, contact Charlotte Yano (yano at cs.cmu.edu) >> >> INDIVIDUAL RATIONALITY AND PARTICIPATION IN LARGE SCALE, >> MULTI-HOSPITAL KIDNEY EXCHANGE >> >> As multi-hospital kidney exchange clearinghouses have grown, the set >> of players has grown from patients and surgeons to include hospitals. >> Hospitals have the option of enrolling only their hard-to-match >> patient-donor pairs, while conducting easily arranged exchanges >> internally. This behavior has already started to be observed. We show >> that the cost of making it individually rational for hospitals to >> participate fully is low in almost every large exchange pool (although >> the worst-case cost is very high), while the cost of failing to >> guarantee individually rational allocations could be large, in terms >> of lost transplants. We also identify an incentive-compatible mechanism. >> >> BIO >> >> Itai Ashlagi is an Assistant Professor at the Sloan School of >> Management. >> He is interested in mechanism design, market design, and game theory. In >> particular, he is interested in both developing and applying economic >> and >> optimization/CS tools for designing better marketplaces. Ashlagi is the >> recipient of the outstanding paper award at the ACM Conference on >> Electronic Commerce 2009. Before coming to MIT, he spent two years as a >> postdoctoral researcher at Harvard Business School. He was also a >> consultant researcher in 2010 for Microsoft Research in New England. He >> graduated from Technion-Israel Institute of Technology in 2008. >> >> >> > -- Dana M. Houston Language Technologies Institute School of Computer Science Carnegie Mellon University 5407 Gates Hillman Complex 5000 Forbes Avenue Pittsburgh, PA 15213 T: (412)268-6591 F: (412)268-6298 From dhouston at cs.cmu.edu Tue Mar 15 08:52:30 2011 From: dhouston at cs.cmu.edu (Dana Houston) Date: Tue, 15 Mar 2011 08:52:30 -0400 Subject: [Intelligence Seminar] CANCELED - March 15, 3:30pm: Presentation by Itai Ashlagi - CANCELED In-Reply-To: <4D7F603C.5050908@cs.cmu.edu> References: <4D77B3E3.4030302@cs.cmu.edu> <4D7E2C79.3030707@cs.cmu.edu> <4D7F603C.5050908@cs.cmu.edu> Message-ID: <4D7F610E.2070008@cs.cmu.edu> Correction: The seminar for TODAY has been canceled. On 3/15/2011 8:49 AM, Dana Houston wrote: > > > The Intelligence Seminar for tomorrow has been canceled. > > > > On 3/14/2011 10:55 AM, Dana Houston wrote: >> >> INTELLIGENCE SEMINAR >>> MARCH 15 AT 3:30PM, IN GHC 4303 >>> >>> SPEAKER: ITAI ASHLAGI (Massachusetts Institute of Technology) >>> Host: Tuomas Sandholm >>> For meetings, contact Charlotte Yano (yano at cs.cmu.edu) >>> >>> INDIVIDUAL RATIONALITY AND PARTICIPATION IN LARGE SCALE, >>> MULTI-HOSPITAL KIDNEY EXCHANGE >>> >>> As multi-hospital kidney exchange clearinghouses have grown, the set >>> of players has grown from patients and surgeons to include hospitals. >>> Hospitals have the option of enrolling only their hard-to-match >>> patient-donor pairs, while conducting easily arranged exchanges >>> internally. This behavior has already started to be observed. We show >>> that the cost of making it individually rational for hospitals to >>> participate fully is low in almost every large exchange pool (although >>> the worst-case cost is very high), while the cost of failing to >>> guarantee individually rational allocations could be large, in terms >>> of lost transplants. We also identify an incentive-compatible >>> mechanism. >>> >>> BIO >>> >>> Itai Ashlagi is an Assistant Professor at the Sloan School of >>> Management. >>> He is interested in mechanism design, market design, and game >>> theory. In >>> particular, he is interested in both developing and applying >>> economic and >>> optimization/CS tools for designing better marketplaces. Ashlagi is the >>> recipient of the outstanding paper award at the ACM Conference on >>> Electronic Commerce 2009. Before coming to MIT, he spent two years as a >>> postdoctoral researcher at Harvard Business School. He was also a >>> consultant researcher in 2010 for Microsoft Research in New England. He >>> graduated from Technion-Israel Institute of Technology in 2008. >>> >>> >>> >> > -- Dana M. Houston Language Technologies Institute School of Computer Science Carnegie Mellon University 5407 Gates Hillman Complex 5000 Forbes Avenue Pittsburgh, PA 15213 T: (412)268-6591 F: (412)268-6298 From dhouston at cs.cmu.edu Fri Mar 18 09:32:02 2011 From: dhouston at cs.cmu.edu (Dana Houston) Date: Fri, 18 Mar 2011 09:32:02 -0400 Subject: [Intelligence Seminar] March 29, 1:30pm: Presentation by Hal Daume Message-ID: <4D835ED2.60407@cs.cmu.edu> INTELLIGENCE SEMINAR MARCH 29 AT 1:30pm, IN GHC 4405 (PLEASE NOTE THE UNUSUAL ROOM AND TIME) SPEAKER: HAL DAUME III (University of Maryland) Host: Noah Smith For meetings, contact Stacey Young (staceyy at cs.cmu.edu) STRUCTURE AND KNOWLEDGE IN NATURAL LANGUAGE PROCESSING Human language exhibits complex structure. To be successful, machine learning approaches to language-related problems must be able to take advantage of this structure. I will discuss several investigations into the relationship between structure and learning, which have led to some surprising conclusions about the role that structure plays in language processing. I will describe some recent efforts related to learning strategies that not only aim to do a good job, but aim to do it quickly. From there, I will consider the question of: where does this structure come from. By taking insights from linguistic typology, I will show that very simple typological information can lead to significant increases in system performance for some simple syntactic problems. Moreover, I will show how this typological information can be mined from raw data. (This talk includes joint work with Dan Klein, John Langford, Percy Liang, Daniel Marcu, and some of my students: Arvind Agarwal, Adam Teichert, and Piyush Rai.) BIO Hal Daume III is an assistant professor in Computer Science at the University of Maryland, College Park. He holds joint appointments in UMIACS and Linguistics. His primary research interest is in understanding how to get human knowledge into a machine learning system in the most efficient way possible. He works primarily in the areas of language (computational linguistics and natural language processing) and machine learning (structured prediction, domain adaptation, and Bayesian inference). He associates himself most with conferences like ACL, ICML, NIPS, and EMNLP, and has over 30 conference papers (one best paper award in ECML/PKDD 2010) and 7 journal papers. He earned his PhD at the University of Southern California with a thesis on structured prediction for language (his advisor was Daniel Marcu). He spent the summer of 2003 working with Eric Brill in the machine learning and applied statistics group at Microsoft Research. Prior to that, he studied math (mostly logic) at Carnegie Mellon University. He still likes math and does not like to use C (instead he uses O'Caml or Haskell). He does not like shoes, but does like activities that are hard on your feet: skiing, badminton, Aikido, and rock climbing. -- Dana M. Houston Language Technologies Institute School of Computer Science Carnegie Mellon University 5407 Gates Hillman Complex 5000 Forbes Avenue Pittsburgh, PA 15213 T: (412)268-6591 F: (412)268-6298 From eugenefink at cmu.edu Sat Mar 26 15:17:19 2011 From: eugenefink at cmu.edu (Eugene Fink) Date: Sat, 26 Mar 2011 15:17:19 -0400 (EDT) Subject: [Intelligence Seminar] March 29, 1:30pm: Presentation by Hal Daume Message-ID: INTELLIGENCE SEMINAR MARCH 29 AT 1:30pm, IN GHC 4405 (PLEASE NOTE THE UNUSUAL ROOM AND TIME) SPEAKER: HAL DAUME III (University of Maryland) Host: Noah Smith For meetings, contact Stacey Young (staceyy at cs.cmu.edu) STRUCTURE AND KNOWLEDGE IN NATURAL LANGUAGE PROCESSING Human language exhibits complex structure. To be successful, machine learning approaches to language-related problems must be able to take advantage of this structure. I will discuss several investigations into the relationship between structure and learning, which have led to some surprising conclusions about the role that structure plays in language processing. I will describe some recent efforts related to learning strategies that not only aim to do a good job, but aim to do it quickly. >From there, I will consider the question of: where does this structure come from. By taking insights from linguistic typology, I will show that very simple typological information can lead to significant increases in system performance for some simple syntactic problems. Moreover, I will show how this typological information can be mined from raw data. (This talk includes joint work with Dan Klein, John Langford, Percy Liang, Daniel Marcu, and some of my students: Arvind Agarwal, Adam Teichert, and Piyush Rai.) BIO Hal Daume III is an assistant professor in Computer Science at the University of Maryland, College Park. He holds joint appointments in UMIACS and Linguistics. His primary research interest is in understanding how to get human knowledge into a machine learning system in the most efficient way possible. He works primarily in the areas of language (computational linguistics and natural language processing) and machine learning (structured prediction, domain adaptation, and Bayesian inference). He associates himself most with conferences like ACL, ICML, NIPS, and EMNLP, and has over 30 conference papers (one best paper award in ECML/PKDD 2010) and 7 journal papers. He earned his PhD at the University of Southern California with a thesis on structured prediction for language (his advisor was Daniel Marcu). He spent the summer of 2003 working with Eric Brill in the machine learning and applied statistics group at Microsoft Research. Prior to that, he studied math (mostly logic) at Carnegie Mellon University. He still likes math and does not like to use C (instead he uses O'Caml or Haskell). He does not like shoes, but does like activities that are hard on your feet: skiing, badminton, Aikido, and rock climbing. -- Dana M. Houston Language Technologies Institute School of Computer Science Carnegie Mellon University 5407 Gates Hillman Complex 5000 Forbes Avenue Pittsburgh, PA 15213 T: (412)268-6591 F: (412)268-6298 From dhouston at cs.cmu.edu Tue Mar 29 08:51:19 2011 From: dhouston at cs.cmu.edu (Dana Houston) Date: Tue, 29 Mar 2011 08:51:19 -0400 Subject: [Intelligence Seminar] March 29, 1:30pm: Presentation by Hal Daume In-Reply-To: References: Message-ID: <4D91D5C7.7040804@cs.cmu.edu> > INTELLIGENCE SEMINAR > MARCH 29 AT 1:30pm, IN GHC 4405 > (PLEASE NOTE THE UNUSUAL ROOM AND TIME) > > SPEAKER: HAL DAUME III (University of Maryland) > Host: Noah Smith > For meetings, contact Stacey Young (staceyy at cs.cmu.edu) > > STRUCTURE AND KNOWLEDGE IN NATURAL LANGUAGE PROCESSING > > Human language exhibits complex structure. To be successful, machine > learning approaches to language-related problems must be able to take > advantage of this structure. I will discuss several investigations into > the relationship between structure and learning, which have led to some > surprising conclusions about the role that structure plays in language > processing. I will describe some recent efforts related to learning > strategies that not only aim to do a good job, but aim to do it quickly. > From there, I will consider the question of: where does this structure > come from. By taking insights from linguistic typology, I will show that > very simple typological information can lead to significant increases in > system performance for some simple syntactic problems. Moreover, I will > show how this typological information can be mined from raw data. > > (This talk includes joint work with Dan Klein, John Langford, Percy > Liang, > Daniel Marcu, and some of my students: Arvind Agarwal, Adam Teichert, and > Piyush Rai.) > > BIO > > Hal Daume III is an assistant professor in Computer Science at the > University of Maryland, College Park. He holds joint appointments in > UMIACS and Linguistics. His primary research interest is in understanding > how to get human knowledge into a machine learning system in the most > efficient way possible. He works primarily in the areas of language > (computational linguistics and natural language processing) and machine > learning (structured prediction, domain adaptation, and Bayesian > inference). He associates himself most with conferences like ACL, ICML, > NIPS, and EMNLP, and has over 30 conference papers (one best paper award > in ECML/PKDD 2010) and 7 journal papers. He earned his PhD at the > University of Southern California with a thesis on structured prediction > for language (his advisor was Daniel Marcu). He spent the summer of 2003 > working with Eric Brill in the machine learning and applied statistics > group at Microsoft Research. Prior to that, he studied math (mostly > logic) > at Carnegie Mellon University. He still likes math and does not like to > use C (instead he uses O'Caml or Haskell). He does not like shoes, but > does like activities that are hard on your feet: skiing, badminton, > Aikido, and rock climbing. > -- Dana M. Houston Language Technologies Institute School of Computer Science Carnegie Mellon University 5407 Gates Hillman Complex 5000 Forbes Avenue Pittsburgh, PA 15213 T: (412)268-6591 F: (412)268-6298 From dhouston at cs.cmu.edu Wed Mar 30 14:30:46 2011 From: dhouston at cs.cmu.edu (Dana Houston) Date: Wed, 30 Mar 2011 14:30:46 -0400 Subject: [Intelligence Seminar] April 5, 3:30pm: Presentation by Elad Yom-Tov Message-ID: <4D9376D6.9020105@cs.cmu.edu> INTELLIGENCE SEMINAR APRIL 5 AT 3:30PM, IN GHC 4303 SPEAKER: ELAD YOM-TOV (Yahoo Research) Host: Donald McGillen For meetings, contact Donald McGillen (mcgillen at yahoo-inc.com) ON THE EFFECT OF SOCIAL AND PHYSICAL DETACHMENT ON INFORMATION NEED The information need of users and the documents which answer it are frequently contingent on the different characteristics of users. This is especially evident during natural disasters, such as earthquakes and violent weather incidents, which create a strong transient information need. In this talk I will describe how the information need of users is affected by their physical and social detachment from the event. Drawing on large-scale data from three major events, I will show that social and physical detachment levels of users are a major influence on their information needs, as manifested by their search engine queries. I will demonstrate that knowing social and physical detachment levels can assist in improving retrieval for two applications: identifying search queries related to events and ranking results in response to event-related queries. Finally, I will discuss the relationship of social and mainstream media to user queries, and show similarities and differences between the three information sources. BIO Elad Yom-Tov is a Senior Research Scientist at Yahoo Research. Before joining Yahoo in 2010, he was with the Machine Learning group at IBM Research Haifa Lab and Rafael. Dr. Yom-Tov received his B.Sc. from Tel-Aviv University and his M.Sc. and Ph.D. from Technion - Israel Institute of Technology. Dr. Yom-Tov has co-authored two books and over 40 publications in top international conferences and journals, and filed over 30 patents (6 of which have been granted so far). Two of his papers won top prizes in their respective venues. His primary research interests are in large-scale Machine Learning, Information Retrieval, and in the past few years, social analysis. -- Dana M. Houston Language Technologies Institute School of Computer Science Carnegie Mellon University 5407 Gates Hillman Complex 5000 Forbes Avenue Pittsburgh, PA 15213 T: (412)268-6591 F: (412)268-6298 From dhouston at cs.cmu.edu Mon Apr 4 09:47:27 2011 From: dhouston at cs.cmu.edu (Dana Houston) Date: Mon, 04 Apr 2011 09:47:27 -0400 Subject: [Intelligence Seminar] April 5, 3:30pm: Presentation by Elad Yom-Tov Message-ID: <4D99CBEF.80801@cs.cmu.edu> INTELLIGENCE SEMINAR APRIL 5 AT 3:30PM, IN GHC 4303 SPEAKER: ELAD YOM-TOV (Yahoo Research) Host: Donald McGillen For meetings, contact Donald McGillen (mcgillen at yahoo-inc.com) ON THE EFFECT OF SOCIAL AND PHYSICAL DETACHMENT ON INFORMATION NEED The information need of users and the documents which answer it are frequently contingent on the different characteristics of users. This is especially evident during natural disasters, such as earthquakes and violent weather incidents, which create a strong transient information need. In this talk I will describe how the information need of users is affected by their physical and social detachment from the event. Drawing on large-scale data from three major events, I will show that social and physical detachment levels of users are a major influence on their information needs, as manifested by their search engine queries. I will demonstrate that knowing social and physical detachment levels can assist in improving retrieval for two applications: identifying search queries related to events and ranking results in response to event-related queries. Finally, I will discuss the relationship of social and mainstream media to user queries, and show similarities and differences between the three information sources. BIO Elad Yom-Tov is a Senior Research Scientist at Yahoo Research. Before joining Yahoo in 2010, he was with the Machine Learning group at IBM Research Haifa Lab and Rafael. Dr. Yom-Tov received his B.Sc. from Tel-Aviv University and his M.Sc. and Ph.D. from Technion - Israel Institute of Technology. Dr. Yom-Tov has co-authored two books and over 40 publications in top international conferences and journals, and filed over 30 patents (6 of which have been granted so far). Two of his papers won top prizes in their respective venues. His primary research interests are in large-scale Machine Learning, Information Retrieval, and in the past few years, social analysis. -- Dana M. Houston Language Technologies Institute School of Computer Science Carnegie Mellon University 5407 Gates Hillman Complex 5000 Forbes Avenue Pittsburgh, PA 15213 T: (412)268-6591 F: (412)268-6298 From dhouston at cs.cmu.edu Tue Apr 5 16:32:59 2011 From: dhouston at cs.cmu.edu (Dana Houston) Date: Tue, 05 Apr 2011 16:32:59 -0400 Subject: [Intelligence Seminar] Found keys Message-ID: <4D9B7C7B.9090900@cs.cmu.edu> If you left a set of keys at the Intelligence Seminar, you may pick them up in GHC 5407 before 5pm. -- Dana M. Houston Language Technologies Institute School of Computer Science Carnegie Mellon University 5407 Gates Hillman Complex 5000 Forbes Avenue Pittsburgh, PA 15213 T: (412)268-6591 F: (412)268-6298 From dhouston at cs.cmu.edu Wed Apr 6 13:32:34 2011 From: dhouston at cs.cmu.edu (Dana Houston) Date: Wed, 06 Apr 2011 13:32:34 -0400 Subject: [Intelligence Seminar] April 12, 3:30pm:, Presentation by Matthew Salganik Message-ID: <4D9CA3B2.5020205@cs.cmu.edu> INTELLIGENCE SEMINAR APRIL 12 AT 3:30PM, IN GHC 4303 SPEAKER: MATTHEW SALGANIK (Princeton University) Host: Burr Settles For meetings, contact Cathy Serventi (serventi at google.com) WIKI SURVEYS: OPEN, ADAPTIVE, AND QUANTIFIABLE SOCIAL DATA COLLECTION Joint work with Karen Levy Research about attitudes and opinions is central to social science and relies on two common methodological approaches: surveys and interviews. While surveys allow researchers to quantify large amounts of information quickly and at a reasonable cost, they are routinely criticized for being "top-down" and rigid. In contrast, interviews allow unanticipated information to "bubble up" directly from respondents, but are slow, expensive, and hard to quantify. Advances in computing technology now enable a hybrid approach, "wiki surveys", that combines the quantifiability of a survey with the openness of an interview. We draw on principles undergirding successful information aggregation projects, such as Wikipedia and the Linux operating system, to propose several theoretical criteria that wiki surveys should satisfy. We then present results fromwww.allourideas.org, a free and open source website that we created, which allows groups all over the world to deploy wiki surveys. To date, over 800 wiki surveys have been created, and they have collected over 30,000 ideas and 2 million votes. We describe some of the methodological challenges involved in collecting and analyzing this type of data, and present case studies of wiki surveys created by the New York City Mayor's Office and the Organization for Economic Co-operation and Development (OECD). The talk concludes a discussion of limitations and how some of these limitations might be overcome with additional research. BIO Matthew Salganik is an Assistant Professor in the Department of Sociology at Princeton University. His interests include social networks, quantitative methods, and web-based social research. One main area of his research has focused on developing network-based statistical methods for studying populations most at risk for HIV/AIDS. A second main area of work has been using the World Wide Web to collect and analyze social data in innovative ways. Salganik's research has been published in journals such as Science, PNAS, Sociological Methodology, and Journal of the American Statistical Association. His papers have won the Outstanding Article Award from the Mathematical Sociology Section of the American Sociological Association and the Outstanding Statistical Application Award from the American Statistical Association. Popular accounts of his work have appeared in the New York Times, Wall Street Journal, Economist, and New Yorker. Salganik's research is funded by the National Science Foundation, National Institutes of Health, Joint United Nations Program for HIV/AIDS (UNAIDS), and Google. -- Dana M. Houston Language Technologies Institute School of Computer Science Carnegie Mellon University 5407 Gates Hillman Complex 5000 Forbes Avenue Pittsburgh, PA 15213 T: (412)268-6591 F: (412)268-6298 From dhouston at cs.cmu.edu Mon Apr 11 12:00:52 2011 From: dhouston at cs.cmu.edu (Dana Houston) Date: Mon, 11 Apr 2011 12:00:52 -0400 Subject: [Intelligence Seminar] April 12, 3:30pm:, Presentation by Matthew Salganik Message-ID: <4DA325B4.30902@cs.cmu.edu> INTELLIGENCE SEMINAR APRIL 12 AT 3:30PM, IN GHC 4303 SPEAKER: MATTHEW SALGANIK (Princeton University) Host: Burr Settles For meetings, contact Cathy Serventi (serventi at google.com) WIKI SURVEYS: OPEN, ADAPTIVE, AND QUANTIFIABLE SOCIAL DATA COLLECTION Joint work with Karen Levy Research about attitudes and opinions is central to social science and relies on two common methodological approaches: surveys and interviews. While surveys allow researchers to quantify large amounts of information quickly and at a reasonable cost, they are routinely criticized for being "top-down" and rigid. In contrast, interviews allow unanticipated information to "bubble up" directly from respondents, but are slow, expensive, and hard to quantify. Advances in computing technology now enable a hybrid approach, "wiki surveys", that combines the quantifiability of a survey with the openness of an interview. We draw on principles undergirding successful information aggregation projects, such as Wikipedia and the Linux operating system, to propose several theoretical criteria that wiki surveys should satisfy. We then present results fromwww.allourideas.org, a free and open source website that we created, which allows groups all over the world to deploy wiki surveys. To date, over 800 wiki surveys have been created, and they have collected over 30,000 ideas and 2 million votes. We describe some of the methodological challenges involved in collecting and analyzing this type of data, and present case studies of wiki surveys created by the New York City Mayor's Office and the Organization for Economic Co-operation and Development (OECD). The talk concludes a discussion of limitations and how some of these limitations might be overcome with additional research. BIO Matthew Salganik is an Assistant Professor in the Department of Sociology at Princeton University. His interests include social networks, quantitative methods, and web-based social research. One main area of his research has focused on developing network-based statistical methods for studying populations most at risk for HIV/AIDS. A second main area of work has been using the World Wide Web to collect and analyze social data in innovative ways. Salganik's research has been published in journals such as Science, PNAS, Sociological Methodology, and Journal of the American Statistical Association. His papers have won the Outstanding Article Award from the Mathematical Sociology Section of the American Sociological Association and the Outstanding Statistical Application Award from the American Statistical Association. Popular accounts of his work have appeared in the New York Times, Wall Street Journal, Economist, and New Yorker. Salganik's research is funded by the National Science Foundation, National Institutes of Health, Joint United Nations Program for HIV/AIDS (UNAIDS), and Google. -- Dana M. Houston Language Technologies Institute School of Computer Science Carnegie Mellon University 5407 Gates Hillman Complex 5000 Forbes Avenue Pittsburgh, PA 15213 T: (412)268-6591 F: (412)268-6298 From dhouston at cs.cmu.edu Mon Apr 18 10:33:59 2011 From: dhouston at cs.cmu.edu (Dana Houston) Date: Mon, 18 Apr 2011 10:33:59 -0400 Subject: [Intelligence Seminar] April 26, 3:30pm:, Presentation by Satinder Singh Message-ID: <4DAC4BD7.6050509@cs.cmu.edu> INTELLIGENCE SEMINAR APRIL 26 AT 3:30PM, IN GHC 4303 SPEAKER: SATINDER SINGH (University of Michigan) Host: Tuomas Sandholm For meetings, contact Charlotte Yano (yano at cs.cmu.edu) THE OPTIMAL REWARD PROBLEM OR WHERE DO REWARDS COME FROM? Joint work with Jonathan Sorg and Richard Lewis Impressive results have been obtained by research approaches to autonomous agents that start with a given reward function and focus on developing theory and algorithms for learning or planning policies that lead to high cumulative reward. In a departure from this work, we recognize that in many situations the starting point is an agent designer with a reward function seeking to build an autonomous agent to act on its behalf. What reward function should the designer build into the autonomous agent? In this new view, setting the parameters (agent's reward function) equal to the given preferences (designer's reward function) implements a preferences-parameters confound. If an agent is bounded, as most agents are in practice, we expect that breaking the preferences-parameters confound would be beneficial. We define the optimal reward problem, that of designing the agent's reward function from among a set of reward functions given a designer's reward function, an agent architecture, and a distribution over environments. The main focus of the talk will be on a discussion of some empirical and theoretical insights obtained by solving the optimal reward problem. BIO Satinder Singh is a Professor of Computer Science and Engineering at the University of Michigan. He is also presently serving as the AI Lab Director. He contributes to the research areas of reinforcement learning, decision-theoretic planning, and computational game theory. -- Dana M. Houston Language Technologies Institute School of Computer Science Carnegie Mellon University 5407 Gates Hillman Complex 5000 Forbes Avenue Pittsburgh, PA 15213 T: (412)268-6591 F: (412)268-6298 From eugenefink at cmu.edu Mon Apr 25 17:18:13 2011 From: eugenefink at cmu.edu (Eugene Fink) Date: Mon, 25 Apr 2011 17:18:13 -0400 (EDT) Subject: [Intelligence Seminar] April 26, 3:30pm: Presentation by Satinder Singh Message-ID: INTELLIGENCE SEMINAR APRIL 26 AT 3:30PM, IN GHC 4303 SPEAKER: SATINDER SINGH (University of Michigan) Host: Tuomas Sandholm For meetings, contact Charlotte Yano (yano at cs.cmu.edu) THE OPTIMAL REWARD PROBLEM OR WHERE DO REWARDS COME FROM? Joint work with Jonathan Sorg and Richard Lewis Impressive results have been obtained by research approaches to autonomous agents that start with a given reward function and focus on developing theory and algorithms for learning or planning policies that lead to high cumulative reward. In a departure from this work, we recognize that in many situations the starting point is an agent designer with a reward function seeking to build an autonomous agent to act on its behalf. What reward function should the designer build into the autonomous agent? In this new view, setting the parameters (agent's reward function) equal to the given preferences (designer's reward function) implements a preferences-parameters confound. If an agent is bounded, as most agents are in practice, we expect that breaking the preferences-parameters confound would be beneficial. We define the optimal reward problem, that of designing the agent's reward function from among a set of reward functions given a designer's reward function, an agent architecture, and a distribution over environments. The main focus of the talk will be on a discussion of some empirical and theoretical insights obtained by solving the optimal reward problem. BIO Satinder Singh is a Professor of Computer Science and Engineering at the University of Michigan. He is also presently serving as the AI Lab Director. He contributes to the research areas of reinforcement learning, decision-theoretic planning, and computational game theory. From dhouston at cs.cmu.edu Thu Jun 2 09:05:07 2011 From: dhouston at cs.cmu.edu (Dana Houston) Date: Thu, 02 Jun 2011 09:05:07 -0400 Subject: [Intelligence Seminar] June 7, 3:30pm:, Presentation by Brian Murphy Message-ID: <4DE78A83.6090804@cs.cmu.edu> INTELLIGENCE SEMINAR JUNE 7 AT 3:30PM, IN GHC 4303 SPEAKER: BRIAN MURPHY (University of Trento) Host: Bob Frederking For meetings, contact Stacey Young (staceyy at cs.cmu.edu) GROUNDING MODELS OF LANGUAGE IN THE BRAIN Over recent decades, linguistics has produced an abundance of theoretical models, while suffering from a lack of empirical robustness. Judgments elicited from native speakers can provide nuanced insights into the psychological states underlying communicative behavior, but are distorted by pervasive cognitive biases. Corpora (large collections of text) have clear advantages of authenticity and size, but are divorced from the communicative context. Recordings of neural activity can provide a happy medium, giving an objective and direct snapshot of the language faculty in action, though the data is noisy and contains confounding activations due to other cognitive processes that typically accompany a language task. In this talk I will describe work we have carried out here at CIMeC, which uses machine learning methods to attempt to distinguish linguistic states and processes in the brain. I will concentrate on decoding lexical meaning, and in particular deal with the many systematic perceptual and performance-based confounds that can make this difficult. BIO Brian Murphy is a postdoctoral fellow at the Centre for Mind/Brain Science, of the University of Trento, Italy. Before moving into research, he gained a degree in engineering and spent 5 years working in software development in Germany and China. He holds an M.Phil. (2001) and Ph.D. (2007) from Trinity College, Dublin, both in computational linguistics, with an emphasis on language semantics and its interaction with sentential syntax. Since coming to Trento, his main topic of research has been word meaning, using machine learning methods to discover conceptual organization both from large language corpora and recordings of neural activity (EEG, MEG and fMRI). -- Dana M. Houston Language Technologies Institute School of Computer Science Carnegie Mellon University 5407 Gates Hillman Complex 5000 Forbes Avenue Pittsburgh, PA 15213 T: (412)268-6591 F: (412)268-6298 From dhouston at cs.cmu.edu Mon Jun 6 08:34:07 2011 From: dhouston at cs.cmu.edu (Dana Houston) Date: Mon, 06 Jun 2011 08:34:07 -0400 Subject: [Intelligence Seminar] June 7, 3:30pm:, Presentation by Brian Murphy In-Reply-To: <4DE78A83.6090804@cs.cmu.edu> References: <4DE78A83.6090804@cs.cmu.edu> Message-ID: <4DECC93F.9010008@cs.cmu.edu> INTELLIGENCE SEMINAR > JUNE 7 AT 3:30PM, IN GHC 4303 > > SPEAKER: BRIAN MURPHY (University of Trento) > Host: Bob Frederking > For meetings, contact Stacey Young (staceyy at cs.cmu.edu) > > GROUNDING MODELS OF LANGUAGE IN THE BRAIN > > Over recent decades, linguistics has produced an abundance of > theoretical models, while suffering from a lack of empirical > robustness. Judgments elicited from native speakers can provide > nuanced insights into the psychological states underlying > communicative behavior, but are distorted by pervasive cognitive > biases. Corpora (large collections of text) have clear advantages of > authenticity and size, but are divorced from the communicative > context. Recordings of neural activity can provide a happy medium, > giving an objective and direct snapshot of the language faculty in > action, though the data is noisy and contains confounding activations > due to other cognitive processes that typically accompany a language > task. In this talk I will describe work we have carried out here at > CIMeC, which uses machine learning methods to attempt to distinguish > linguistic states and processes in the brain. I will concentrate on > decoding lexical meaning, and in particular deal with the many > systematic perceptual and performance-based confounds that can make > this difficult. > > BIO > > Brian Murphy is a postdoctoral fellow at the Centre for Mind/Brain > Science, of the University of Trento, Italy. Before moving into > research, he gained a degree in engineering and spent 5 years working > in software development in Germany and China. He holds an M.Phil. > (2001) and Ph.D. (2007) from Trinity College, Dublin, both in > computational linguistics, with an emphasis on language semantics and > its interaction with sentential syntax. Since coming to Trento, his > main topic of research has been word meaning, using machine learning > methods to discover conceptual organization both from large language > corpora and recordings of neural activity (EEG, MEG and fMRI). > -- Dana M. Houston Language Technologies Institute School of Computer Science Carnegie Mellon University 5407 Gates Hillman Complex 5000 Forbes Avenue Pittsburgh, PA 15213 T: (412)268-6591 F: (412)268-6298 From dhouston at cs.cmu.edu Wed Sep 14 15:55:45 2011 From: dhouston at cs.cmu.edu (Dana Houston) Date: Wed, 14 Sep 2011 15:55:45 -0400 Subject: [Intelligence Seminar] Sept. 20, 3:30pm:, Presentation by Daniel Tunkelang In-Reply-To: References: Message-ID: <4E7106C1.8060309@cs.cmu.edu> INTELLIGENCE SEMINAR SEPTEMBER 20 AT 3:30PM, IN GHC 4303 SPEAKER: DANIEL TUNKELANG (LinkedIn) Host: Eugene Fink For meetings, contact Dana Houston (dhouston at cs.cmu.edu) KEEPING IT PROFESSIONAL: RELEVANCE, RECOMMENDATIONS, AND REPUTATION AT LINKEDIN LinkedIn operates the world's largest professional network on the Internet with more than 120 million members in over 200 countries. In order to connect its users to the people, opportunities, and content that best advance their careers, LinkedIn has developed a variety of algorithms that surface relevant content, offer personalized recommendations, and establish topic-sensitive reputation -- all at a massive scale. In this talk, I will discuss some of the most challenging technical problems we face at LinkedIn, and the approaches we are taking to address them. BIO Daniel Tunkelang oversees the data science team at LinkedIn, which analyzes terabytes of data to produce products and insights that serve LinkedIn's members. Prior to LinkedIn, Daniel led a local search quality team at Google. Daniel was a founding employee and Chief Scientist of Endeca, a leader in enterprise search and business intelligence that pioneered the use of guided navigation in search applications. He has authored eight patents, written a textbook on faceted search, created the annual workshop on human-computer interaction and information retrieval (HCIR), and participated in the premier research conferences on information retrieval, knowledge management, databases, and data mining (SIGIR, CIKM, SIGMOD, SIAM Data Mining). Daniel holds a Ph.D. in Computer Science from CMU, as well as B.S. and M.S. degrees from MIT. -- Dana M. Houston Language Technologies Institute School of Computer Science Carnegie Mellon University 5405 Gates Hillman Complex 5000 Forbes Avenue Pittsburgh, PA 15213 T: (412)268-4717 F: (412)268-6298 From dhouston at cs.cmu.edu Tue Sep 20 09:20:46 2011 From: dhouston at cs.cmu.edu (Dana Houston) Date: Tue, 20 Sep 2011 09:20:46 -0400 Subject: [Intelligence Seminar] Sept. 20, 3:30pm:, Presentation by Daniel Tunkelang In-Reply-To: <4E7106C1.8060309@cs.cmu.edu> References: <4E7106C1.8060309@cs.cmu.edu> Message-ID: <4E78932E.8040702@cs.cmu.edu> > > INTELLIGENCE SEMINAR > SEPTEMBER 20 AT 3:30PM, IN GHC 4303 > > SPEAKER: DANIEL TUNKELANG (LinkedIn) > Host: Eugene Fink > For meetings, contact Dana Houston (dhouston at cs.cmu.edu) > > KEEPING IT PROFESSIONAL: > RELEVANCE, RECOMMENDATIONS, AND REPUTATION AT LINKEDIN > > LinkedIn operates the world's largest professional network on the > Internet with more than 120 million members in over 200 countries. In > order to connect its users to the people, opportunities, and content > that best advance their careers, LinkedIn has developed a variety of > algorithms that surface relevant content, offer personalized > recommendations, and establish topic-sensitive reputation -- all at a > massive scale. In this talk, I will discuss some of the most > challenging technical problems we face at LinkedIn, and the approaches > we are taking to address them. > > BIO > > Daniel Tunkelang oversees the data science team at LinkedIn, which > analyzes terabytes of data to produce products and insights that serve > LinkedIn's members. Prior to LinkedIn, Daniel led a local search > quality team at Google. Daniel was a founding employee and Chief > Scientist of Endeca, a leader in enterprise search and business > intelligence that pioneered the use of guided navigation in search > applications. He has authored eight patents, written a textbook on > faceted search, created the annual workshop on human-computer > interaction and information retrieval (HCIR), and participated in the > premier research conferences on information retrieval, knowledge > management, databases, and data mining (SIGIR, CIKM, SIGMOD, SIAM Data > Mining). Daniel holds a Ph.D. in Computer Science from CMU, as well as > B.S. and M.S. degrees from MIT. > > -- Dana M. Houston Language Technologies Institute School of Computer Science Carnegie Mellon University 5405 Gates Hillman Complex 5000 Forbes Avenue Pittsburgh, PA 15213 T: (412)268-4717 F: (412)268-6298 From dhouston at cs.cmu.edu Mon Oct 24 15:38:44 2011 From: dhouston at cs.cmu.edu (Dana Houston) Date: Mon, 24 Oct 2011 15:38:44 -0400 Subject: [Intelligence Seminar] Nov. 1, 3:30pm:, Presentation by David Pennock Message-ID: <4EA5BEC4.8090807@cs.cmu.edu> INTELLIGENCE SEMINAR NOVEMBER 1 AT 3:30PM, IN GHC 4303 SPEAKER: DAVID PENNOCK (Yahoo Research) Host: Ariel Procaccia For meetings, contact Gayle Bishop (gayle at cs.cmu.edu) Mechanism design, or "inverse game theory" is an engineering arm of social science. I will discuss some of our work designing mechanisms to acquire and aggregate information with the goal of making predictions. I will focus on the engineering questions: How do they work and why? What factors and goals are most important in their design? Two somewhat nonstandard objectives are important for good prediction mechanisms: liquidity and expressiveness. Liquidity ensures that agents can be compensated for their information at any time, even when few others are around. I will describe our designs for several automated market maker algorithms that provide desired levels of liquidity. An expressive mechanism offers agents flexibility in how they communicate information; at the extreme, agents can provide any information they have in any form they like. I will discuss our work on combinatorial prediction markets that take expressiveness to the extreme. BIO David Pennock is a Principal Research Scientist at Yahoo! Research in New York City, where he leads a group focused on algorithmic economics. He has over sixty academic publications relating to computational issues in electronic commerce and the web, including papers in PNAS, Science, IEEE Computer, Theoretical Computer Science, Algorithmica, Electronic Commerce Research, Electronic Markets, AAAI, EC, WWW, KDD, UAI, SIGIR, ICML, NIPS, INFOCOM, SAINT, ACM SIGCSE, and VLDB. He has authored two patents and ten patent applications. In 2005, he was named to MIT Technology Review's list of 35 top technology innovators under age 35. Prior to his current position at Yahoo!, Dr. Pennock worked as a research scientist at NEC Laboratories America, a research intern at Microsoft Research, and in 2001 served as an adjunct professor at Pennsylvania State University. He received a Ph.D. in Computer Science from the University of Michigan, an M.S. in Computer Science from Duke University, and a B.S. in Physics from Duke. Dr. Pennock's work has been featured in Discover Magazine, New Scientist, CNN, the New York Times, the Economist, Surowieckias "The Wisdom of Crowds", and several other publications. -- Dana M. Houston Language Technologies Institute School of Computer Science Carnegie Mellon University 5405 Gates Hillman Complex 5000 Forbes Avenue Pittsburgh, PA 15213 T: (412)268-4717 F: (412)268-6298 From dhouston at cs.cmu.edu Mon Oct 31 11:23:50 2011 From: dhouston at cs.cmu.edu (Dana Houston) Date: Mon, 31 Oct 2011 11:23:50 -0400 Subject: [Intelligence Seminar] Nov. 1, 3:30pm:, Presentation by David Pennock Message-ID: <4EAEBD86.1050604@cs.cmu.edu> INTELLIGENCE SEMINAR NOVEMBER 1 AT 3:30PM, IN GHC 4303 SPEAKER: DAVID PENNOCK (Yahoo Research) Host: Ariel Procaccia For meetings, contact Gayle Bishop (gayle at cs.cmu.edu) MECHANISM DESIGN FOR PREDICTION Mechanism design, or "inverse game theory" is an engineering arm of social science. I will discuss some of our work designing mechanisms to acquire and aggregate information with the goal of making predictions. I will focus on the engineering questions: How do they work and why? What factors and goals are most important in their design? Two somewhat nonstandard objectives are important for good prediction mechanisms: liquidity and expressiveness. Liquidity ensures that agents can be compensated for their information at any time, even when few others are around. I will describe our designs for several automated market maker algorithms that provide desired levels of liquidity. An expressive mechanism offers agents flexibility in how they communicate information; at the extreme, agents can provide any information they have in any form they like. I will discuss our work on combinatorial prediction markets that take expressiveness to the extreme. BIO David Pennock is a Principal Research Scientist at Yahoo! Research in New York City, where he leads a group focused on algorithmic economics. He has over sixty academic publications relating to computational issues in electronic commerce and the web, including papers in PNAS, Science, IEEE Computer, Theoretical Computer Science, Algorithmica, Electronic Commerce Research, Electronic Markets, AAAI, EC, WWW, KDD, UAI, SIGIR, ICML, NIPS, INFOCOM, SAINT, ACM SIGCSE, and VLDB. He has authored two patents and ten patent applications. In 2005, he was named to MIT Technology Review's list of 35 top technology innovators under age 35. Prior to his current position at Yahoo!, Dr. Pennock worked as a research scientist at NEC Laboratories America, a research intern at Microsoft Research, and in 2001 served as an adjunct professor at Pennsylvania State University. He received a Ph.D. in Computer Science from the University of Michigan, an M.S. in Computer Science from Duke University, and a B.S. in Physics from Duke. Dr. Pennock's work has been featured in Discover Magazine, New Scientist, CNN, the New York Times, the Economist, Surowieckias "The Wisdom of Crowds", and several other publications. -- Dana M. Houston Language Technologies Institute School of Computer Science Carnegie Mellon University 5405 Gates Hillman Complex 5000 Forbes Avenue Pittsburgh, PA 15213 T: (412)268-4717 F: (412)268-6298 From dhouston at cs.cmu.edu Wed Nov 2 08:32:53 2011 From: dhouston at cs.cmu.edu (Dana Houston) Date: Wed, 02 Nov 2011 08:32:53 -0400 Subject: [Intelligence Seminar] Nov. 8, 2:00pm:, Presentation by Vincent Conitzer Message-ID: <4EB13875.4020200@cs.cmu.edu> INTELLIGENCE SEMINAR NOVEMBER 8 AT 2:00PM, IN GHC 4405 (PLEASE NOTE THE UNUSUAL ROOM AND TIME) SPEAKER: VINCENT CONITZER (Duke University) Host: Ariel Procaccia For meetings, contact Gayle Bishop (gayle at cs.cmu.edu) MECHANISM DESIGN FOR PREDICTION Joint work with Dmytro Korzhyk, Joshua Letchford, Kamesh Munagala, Ronald Parr (Duke); Manish Jain, Zhengyu Yin, Milind Tambe (USC); Christopher Kiekintveld (UT El Paso); Ondrej Vanek, Michal Pechoucek (Czech Technical University); Tuomas Sandholm (CMU) Algorithms for computing game-theoretic solutions are now deployed in real-world security domains, notably air travel. These applications raise some hard questions. How do we deal with the equilibrium selection problem? How is the temporal and informational structure of the game best modeled? What assumptions can we reasonably make about the utility functions of the attacker and the defender? And, last but not least, can we make all these modeling decisions in a way that allows us to scale to realistic instances? I will present our ongoing work on answering these questions. BIO Vincent Conitzer is the Sally Dalton Robinson Professor of Computer Science and Professor of Economics at Duke University, and received his Ph.D. from CMU CSD in 2006, advised by Tuomas Sandholm. His research focuses on computational aspects of microeconomic theory, in particular game theory, mechanism design, voting/social choice, and auctions. He recently received the IJCAI Computers and Thought Award, which is awarded to outstanding young scientists in artificial intelligence. -- Dana M. Houston Language Technologies Institute School of Computer Science Carnegie Mellon University 5405 Gates Hillman Complex 5000 Forbes Avenue Pittsburgh, PA 15213 T: (412)268-4717 F: (412)268-6298 From dhouston at cs.cmu.edu Wed Nov 2 08:36:50 2011 From: dhouston at cs.cmu.edu (Dana Houston) Date: Wed, 02 Nov 2011 08:36:50 -0400 Subject: [Intelligence Seminar] Nov. 8, 2:00pm:, Presentation by Vincent Conitzer (CORRECTION) Message-ID: <4EB13962.5070702@cs.cmu.edu> (CORRECTION) INTELLIGENCE SEMINAR NOVEMBER 8 AT 2:00PM, IN GHC 4405 (PLEASE NOTE THE UNUSUAL ROOM AND TIME) SPEAKER: VINCENT CONITZER (Duke University) Host: Ariel Procaccia For meetings, contact Gayle Bishop (gayle at cs.cmu.edu) COMPUTING GAME-THEORETIC SOLUTIONS FOR SECURITY Joint work with Dmytro Korzhyk, Joshua Letchford, Kamesh Munagala, Ronald Parr (Duke); Manish Jain, Zhengyu Yin, Milind Tambe (USC); Christopher Kiekintveld (UT El Paso); Ondrej Vanek, Michal Pechoucek (Czech Technical University); Tuomas Sandholm (CMU) Algorithms for computing game-theoretic solutions are now deployed in real-world security domains, notably air travel. These applications raise some hard questions. How do we deal with the equilibrium selection problem? How is the temporal and informational structure of the game best modeled? What assumptions can we reasonably make about the utility functions of the attacker and the defender? And, last but not least, can we make all these modeling decisions in a way that allows us to scale to realistic instances? I will present our ongoing work on answering these questions. BIO Vincent Conitzer is the Sally Dalton Robinson Professor of Computer Science and Professor of Economics at Duke University, and received his Ph.D. from CMU CSD in 2006, advised by Tuomas Sandholm. His research focuses on computational aspects of microeconomic theory, in particular game theory, mechanism design, voting/social choice, and auctions. He recently received the IJCAI Computers and Thought Award, which is awarded to outstanding young scientists in artificial intelligence. -- Dana M. Houston Language Technologies Institute School of Computer Science Carnegie Mellon University 5405 Gates Hillman Complex 5000 Forbes Avenue Pittsburgh, PA 15213 T: (412)268-4717 F: (412)268-6298 From dhouston at cs.cmu.edu Mon Nov 7 08:37:02 2011 From: dhouston at cs.cmu.edu (Dana Houston) Date: Mon, 07 Nov 2011 08:37:02 -0500 Subject: [Intelligence Seminar] Nov. 8, 2:00pm:, Presentation by Vincent Conitzer Message-ID: <4EB7DEFE.1060809@cs.cmu.edu> INTELLIGENCE SEMINAR NOVEMBER 8 AT 2:00PM, IN GHC 4405 (PLEASE NOTE THE UNUSUAL ROOM AND TIME) SPEAKER: VINCENT CONITZER (Duke University) Host: Ariel Procaccia For meetings, contact Gayle Bishop (gayle at cs.cmu.edu) COMPUTING GAME-THEORETIC SOLUTIONS FOR SECURITY Joint work with Dmytro Korzhyk, Joshua Letchford, Kamesh Munagala, Ronald Parr (Duke); Manish Jain, Zhengyu Yin, Milind Tambe (USC); Christopher Kiekintveld (UT El Paso); Ondrej Vanek, Michal Pechoucek (Czech Technical University); Tuomas Sandholm (CMU) Algorithms for computing game-theoretic solutions are now deployed in real-world security domains, notably air travel. These applications raise some hard questions. How do we deal with the equilibrium selection problem? How is the temporal and informational structure of the game best modeled? What assumptions can we reasonably make about the utility functions of the attacker and the defender? And, last but not least, can we make all these modeling decisions in a way that allows us to scale to realistic instances? I will present our ongoing work on answering these questions. BIO Vincent Conitzer is the Sally Dalton Robinson Professor of Computer Science and Professor of Economics at Duke University, and received his Ph.D. from CMU CSD in 2006, advised by Tuomas Sandholm. His research focuses on computational aspects of microeconomic theory, in particular game theory, mechanism design, voting/social choice, and auctions. He recently received the IJCAI Computers and Thought Award, which is awarded to outstanding young scientists in artificial intelligence. -- Dana M. Houston Language Technologies Institute School of Computer Science Carnegie Mellon University 5405 Gates Hillman Complex 5000 Forbes Avenue Pittsburgh, PA 15213 T: (412)268-4717 F: (412)268-6298 From dhouston at cs.cmu.edu Tue Nov 8 14:59:31 2011 From: dhouston at cs.cmu.edu (Dana Houston) Date: Tue, 08 Nov 2011 14:59:31 -0500 Subject: [Intelligence Seminar] Nov. 15, 3:30pm:, Presentation by Mark Dredze Message-ID: <4EB98A23.3050509@cs.cmu.edu> INTELLIGENCE SEMINAR NOVEMBER 15 AT 3:30PM, IN GHC 4303 SPEAKER: MARK DREDZE (Johns Hopkins University) Host: Carolyn Penstein Rose For meetings, contact Dana Houston (dhouston at cs.cmu.edu) TOPIC MODELS FOR MINING PUBLIC HEALTH INFORMATION FROM TWITTER Twitter and other social media sites contain a wealth of information about populations and have been used to track sentiment towards products, measure political attitudes, and study social linguistics. In this talk, we investigate the potential for Twitter to impact public health research. Specifically, we consider population surveillance, a major focus of public health that typically depends on clinical encounters with health professionals to collect patient data. Individual users often broadcast salient health information, such as "sick with this flu fever taking over my body ughhhh time for tylenol", which indicates that not only does this person have the flu, but also a fever and is self-medicating with tylenol. Aggregating such content across millions of users could provide information about numerous aspects of illnesses in the population. In this work we present the Ailment Topic Aspect Model (ATAM), a new Bayesian graphical model for Twitter that associates symptoms, treatments, and general words with diseases (ailments). When applied to 1.6 million health-related tweets, ATAM discovers descriptions of diseases in terms of collections of words (symptoms and treatments) and partitions messages based on the referenced disease. The model discovers diseases corresponding to influenza, infections, obesity, insomnia, and several others. Furthermore, we demonstrate the effectiveness of this model at several tasks: tracking illnesses over times (syndromic surveillance), measuring behavioral risk factors, localizing illnesses by geographic region, and analyzing symptoms and medication usage. We show quantitative correlations with public health data and qualitative evaluations of model output. Our results suggest that Twitter has broad applicability for public health research. BIO Mark Dredze is an Assistant Research Professor in Computer Science at Johns Hopkins University, as well as a member of the Center for Language and Speech Processing and the Human Language Technology Center of Excellence. His research in natural language processing and machine learning has focused on graphical models, semi-supervised learning, information extraction, large-scale learning, speech processing, and health informatics. He obtained his PhD from the University of Pennsylvania in 2009. -- Dana M. Houston Language Technologies Institute School of Computer Science Carnegie Mellon University 5405 Gates Hillman Complex 5000 Forbes Avenue Pittsburgh, PA 15213 T: (412)268-4717 F: (412)268-6298 From dhouston at cs.cmu.edu Mon Nov 14 12:17:08 2011 From: dhouston at cs.cmu.edu (Dana Houston) Date: Mon, 14 Nov 2011 12:17:08 -0500 Subject: [Intelligence Seminar] Nov. 15, 3:30pm:, Presentation by Mark Dredze In-Reply-To: <4EB98A23.3050509@cs.cmu.edu> References: <4EB98A23.3050509@cs.cmu.edu> Message-ID: <4EC14D14.7010606@cs.cmu.edu> > INTELLIGENCE SEMINAR > NOVEMBER 15 AT 3:30PM, IN GHC 4303 > > SPEAKER: MARK DREDZE (Johns Hopkins University) > Host: Carolyn Penstein Rose > For meetings, contact Dana Houston (dhouston at cs.cmu.edu) > > TOPIC MODELS FOR MINING PUBLIC HEALTH INFORMATION FROM TWITTER > > Twitter and other social media sites contain a wealth of information > about > populations and have been used to track sentiment towards products, > measure > political attitudes, and study social linguistics. In this talk, we > investigate the potential for Twitter to impact public health research. > Specifically, we consider population surveillance, a major focus of > public > health that typically depends on clinical encounters with health > professionals to collect patient data. Individual users often broadcast > salient health information, such as "sick with this flu fever taking over > my body ughhhh time for tylenol", which indicates that not only does this > person have the flu, but also a fever and is self-medicating with > tylenol. > Aggregating such content across millions of users could provide > information about numerous aspects of illnesses in the population. > > In this work we present the Ailment Topic Aspect Model (ATAM), a new > Bayesian graphical model for Twitter that associates symptoms, > treatments, > and general words with diseases (ailments). When applied to 1.6 million > health-related tweets, ATAM discovers descriptions of diseases in > terms of > collections of words (symptoms and treatments) and partitions messages > based on the referenced disease. The model discovers diseases > corresponding to influenza, infections, obesity, insomnia, and several > others. Furthermore, we demonstrate the effectiveness of this model at > several tasks: tracking illnesses over times (syndromic surveillance), > measuring behavioral risk factors, localizing illnesses by geographic > region, and analyzing symptoms and medication usage. We show quantitative > correlations with public health data and qualitative evaluations of model > output. Our results suggest that Twitter has broad applicability for > public health research. > > BIO > > Mark Dredze is an Assistant Research Professor in Computer Science at > Johns Hopkins University, as well as a member of the Center for Language > and Speech Processing and the Human Language Technology Center of > Excellence. His research in natural language processing and machine > learning has focused on graphical models, semi-supervised learning, > information extraction, large-scale learning, speech processing, and > health informatics. He obtained his PhD from the University of > Pennsylvania in 2009. > -- Dana M. Houston Language Technologies Institute School of Computer Science Carnegie Mellon University 5405 Gates Hillman Complex 5000 Forbes Avenue Pittsburgh, PA 15213 T: (412)268-4717 F: (412)268-6298 From dhouston at cs.cmu.edu Tue Nov 22 14:48:14 2011 From: dhouston at cs.cmu.edu (Dana Houston) Date: Tue, 22 Nov 2011 14:48:14 -0500 Subject: [Intelligence Seminar] Nov. 29, 3:30pm:, Presentation by Gerry Tesauro (IBM Research) Message-ID: <4ECBFC7E.6030005@cs.cmu.edu> INTELLIGENCE SEMINAR NOVEMBER 29 AT 3:30PM, IN GHC 4303 SPEAKER: GERRY TESAURO (IBM Research) Host: Burr Settles For meetings, contact Dana Houston (dhouston at cs.cmu.edu) HOW WATSON LEARNS SUPERHUMAN JEOPARDY! STRATEGIES Major advances in Question Answering technology were needed for Watson to play Jeopardy! at championship level--the show requires rapid-fire answers to challenging natural language questions, broad general knowledge, high precision, and accurate confidence estimates. In addition, Jeopardy! features four types of decision making carrying great strategic importance: (1) selecting the next clue when in control of the board; (2) deciding whether to attempt to buzz in; (3) wagering on Daily Doubles; (4) wagering in Final Jeopardy. This talk describes how Watson makes the above decisions using innovative quantitative methods that, in principle, maximize Watson's overall winning chances. We first describe our development of faithful simulation models of human contestants and the Jeopardy! game environment. We then present specific learning/optimization methods used in each strategy algorithm: these methods span a range of popular AI research topics, including Bayesian inference, game theory, Dynamic Programming, Reinforcement Learning, and real-time "rollouts." Application of these methods yielded superhuman game strategies for Watson that significantly enhanced its overall competitive record. BIO Gerald Tesauro is a Research Staff Member at IBM's TJ Watson Research Center. He is best known for developing TD-Gammon, a self-teaching neural network that learned to play backgammon at human world championship level. He has also worked on theoretical and applied machine learning in a wide variety of other settings, including multi-agent learning, dimensionality reduction, computer virus recognition, computer chess (Deep Blue), intelligent e-commerce agents, and autonomic computing. Dr. Tesauro received BS and PhD degrees in physics from University of Maryland and Princeton University, respectively. -- Dana M. Houston Language Technologies Institute School of Computer Science Carnegie Mellon University 5405 Gates Hillman Complex 5000 Forbes Avenue Pittsburgh, PA 15213 T: (412)268-4717 F: (412)268-6298 From dhouston at cs.cmu.edu Mon Nov 28 08:40:07 2011 From: dhouston at cs.cmu.edu (Dana Houston) Date: Mon, 28 Nov 2011 08:40:07 -0500 Subject: [Intelligence Seminar] Nov. 29, 3:30pm:, Presentation by Gerry Tesauro (IBM Research) In-Reply-To: <4ECC0C5A.5000102@cs.cmu.edu> References: <4ECBFC7E.6030005@cs.cmu.edu> <4ECC0C5A.5000102@cs.cmu.edu> Message-ID: <4ED38F37.8010204@cs.cmu.edu> Please join us for the next Intelligence Seminar! Tuesday, November 29, 2011 3:30pm Gates Building 4303 Host: Burr Settles, for appointments please contact Dana Houston (dhouston at cs.cmu.edu ) Speaker: Gerry Tesauro, IBM Research Title: How Watson Learns Superhuman Jeopardy! Strategies Abstract: Major advances in Question Answering technology were needed for Watson to play Jeopardy! at championship level--the show requires rapid-fire answers to challenging natural language questions, broad general knowledge, high precision, and accurate confidence estimates. In addition, Jeopardy! features four types of decision making carrying great strategic importance: (1) selecting the next clue when in control of the board; (2) deciding whether to attempt to buzz in; (3) wagering on Daily Doubles; (4) wagering in Final Jeopardy. This talk describes how Watson makes the above decisions using innovative quantitative methods that, in principle, maximize Watson's overall winning chances. We first describe our development of faithful simulation models of human contestants and the Jeopardy! game environment. We then present specific learning/optimization methods used in each strategy algorithm: these methods span a range of popular AI research topics, including Bayesian inference, game theory, Dynamic Programming, Reinforcement Learning, and real-time "rollouts." Application of these methods yielded superhuman game strategies for Watson that significantly enhanced its overall competitive record. Joint work with David Gondek, Jon Lenchner, James Fan, and John Prager Gerald Tesauro is a Research Staff Member at IBM's TJ Watson Research Center. He is best known for developing TD-Gammon, a self-teaching neural network that learned to play backgammon at human world championship level. He has also worked on theoretical and applied machine learning in a wide variety of other settings, including multi-agent learning, dimensionality reduction, computer virus recognition, computer chess (Deep Blue), intelligent e-commerce agents, and autonomic computing. Dr. Tesauro received BS and PhD degrees in physics from University of Maryland and Princeton University, respectively. -------------- next part -------------- An HTML attachment was scrubbed... URL: http://mailman.srv.cs.cmu.edu/pipermail/intelligence-seminar-announce/attachments/20111128/51c597b5/attachment.html