From macglashan at tti-c.org Mon Aug 3 10:07:58 2009 From: macglashan at tti-c.org (Julia MacGlashan) Date: Mon Aug 3 10:08:14 2009 Subject: [TTIC Colloquium] TTI-C Talk: Sivan Sabato, Hebrew University Message-ID: <43D5489227904FD68ECE8D8A2BFAB20E@jmacglDPLFYD1> When: Tuesday, Aug 4 @ 11:00am Where: 6045 S Kenwood Ave, TTI-C Conference Room #526 (5th Floor) Who: Sivan Sabato (Hebrew University) Title: Reducing Label Complexity using Multiple Instance Learning A number of learning problems can be cast in a supervised learning setting in which the main cost of learning is the number of training labels, and one can obtain a single label for a bag of examples, where the label indicates only if a positive example exists in the bag. We show that in this setting it can be worthwhile to let a teacher label a training sample of bags instead of a training sample of individual examples. This paradigm is analyzed as a special case of Multiple Instance Learning, showing analytically how to select the bag size as a function of the problem parameters, and proving that if the original labels are distributed unevenly, the number of required labels drops considerably when learning from bags. PMIL, A learning algorithm for finding a low-error separating hyperplane from bags, is presented. Experiments on both synthetic and real data sets demonstrate that the method indeed reduces the required number of labels while obtaining the same classification error. If time permits we will also discuss some more general results pertaining to the Multiple Instance Learning problem. Joint Work with Nati Srebro and Naftali Tishby. Contact: Nati Srebro, TTI-C nati@tti-c.org 834-7493 -------------- next part -------------- An HTML attachment was scrubbed... URL: http://ttic.uchicago.edu/pipermail/colloquium/attachments/20090803/80845128/attachment-0001.htm From macglashan at tti-c.org Mon Aug 3 10:09:51 2009 From: macglashan at tti-c.org (Julia MacGlashan) Date: Mon Aug 3 10:10:04 2009 Subject: [TTIC Colloquium] TTI-C Colloquium TODAY: Jacob Goldberger (Bar-Ilan University, Israel) References: <9BFA4FFE1ACE407581765CA2326E1547@jmacglDPLFYD1> Message-ID: <38F9F5E261914E46B397B52D2E91DFD3@jmacglDPLFYD1> REMINDER When: TODAY- Monday, Aug 3 @ 2:00pm Where: 6045 S Kenwood Ave, TTI-C Conference Room #526 (5th Floor) Who: Jacob Goldberger (Bar-Ilan University, Israel) Title: An effective belief-propagation algorithm for constrained linear least-squares problems In this talk we propose a new algorithm for the linear least squares problem where the unknown variables are constrained to be in a finite set. The factor graph that corresponds to this problem is very loopy; in fact, it is a complete graph. Hence, applying the Belief Propagation algorithm yields very poor results. The algorithm described here is based on an optimal tree approximation of the Gaussian density of the unconstrained linear system. It is shown that even though the approximation is not directly applied to the exact discrete distribution, applying the BP algorithm to the modified loop-free factor graph outperforms current methods in terms of both performance and complexity. Joint work with Amir Leshem. Contact: Nati Srebro, TTI-C nati@tti-c.org 834-7493 Speaker Schedule: http://www.tti-c.org/colloquium.php -------------- next part -------------- A non-text attachment was scrubbed... Name: winmail.dat Type: application/ms-tnef Size: 12766 bytes Desc: not available Url : http://ttic.uchicago.edu/pipermail/colloquium/attachments/20090803/5f292f86/winmail.bin From nati at uchicago.edu Tue Aug 4 10:52:03 2009 From: nati at uchicago.edu (Nathan Srebro) Date: Tue Aug 4 10:52:27 2009 Subject: [TTIC Colloquium] NOW: [Colloquium] TTI-C Talk: Sivan Sabato, Hebrew University Message-ID: <154084920908040852y6b70ba30nec33a9e1c1ed046f@mail.gmail.com> ---------- Forwarded message ---------- From: Julia MacGlashan Date: 2009/8/3 Subject: [Colloquium] TTI-C Talk: Sivan Sabato, Hebrew University To: colloquium@nagoya.uchicago.edu, colloquium@cs.uchicago.edu When: ??????????? Tuesday, Aug 4 @ 11:00am Where:??????????? 6045 S Kenwood Ave, TTI-C Conference Room #526 (5th Floor) Who:?????????????? Sivan Sabato (Hebrew University) Title:??? ??????????? Reducing Label Complexity using Multiple Instance Learning A number of learning problems can be cast in a supervised learning setting in which the main cost of learning is the number of training labels, and one can obtain a single label for a bag of examples, where the label indicates only if a positive example exists in the bag. We show that in this setting it can be worthwhile to let a teacher label a training sample of bags instead of a training sample of individual examples. This paradigm is analyzed as a special case of Multiple Instance Learning, showing analytically how to select the bag size as a function of the problem parameters, and proving that if the original labels are distributed unevenly, the number of required labels drops considerably when learning from bags. PMIL, A learning algorithm for finding a low-error separating hyperplane from bags, is presented. Experiments on both synthetic and real data sets demonstrate that the method indeed reduces the required number of labels while obtaining the same classification error. If time permits we will also discuss some more general results pertaining to the Multiple Instance Learning problem. Joint Work with Nati Srebro and Naftali Tishby. Contact:????????? Nati Srebro, TTI-C?????? nati@tti-c.org????????????? 834-7493 _______________________________________________ Colloquium mailing list ?- ?Colloquium@mailman.cs.uchicago.edu https://mailman.cs.uchicago.edu/mailman/listinfo/colloquium From macglashan at tti-c.org Tue Aug 4 11:38:40 2009 From: macglashan at tti-c.org (Julia MacGlashan) Date: Tue Aug 4 11:38:52 2009 Subject: [TTIC Colloquium] Sivan Sabato Talk Rescheduled to 2pm TODAY Message-ID: <7803EA9976F544EDA7D256F33580D9FE@jmacglDPLFYD1> ****This talk has been rescheduled to 2:00pm today**** _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ When: Tuesday, Aug 4 @ 11:00am Where: 6045 S Kenwood Ave, TTI-C Conference Room #526 (5th Floor) Who: Sivan Sabato (Hebrew University) Title: Reducing Label Complexity using Multiple Instance Learning A number of learning problems can be cast in a supervised learning setting in which the main cost of learning is the number of training labels, and one can obtain a single label for a bag of examples, where the label indicates only if a positive example exists in the bag. We show that in this setting it can be worthwhile to let a teacher label a training sample of bags instead of a training sample of individual examples. This paradigm is analyzed as a special case of Multiple Instance Learning, showing analytically how to select the bag size as a function of the problem parameters, and proving that if the original labels are distributed unevenly, the number of required labels drops considerably when learning from bags. PMIL, A learning algorithm for finding a low-error separating hyperplane from bags, is presented. Experiments on both synthetic and real data sets demonstrate that the method indeed reduces the required number of labels while obtaining the same classification error. If time permits we will also discuss some more general results pertaining to the Multiple Instance Learning problem. Joint Work with Nati Srebro and Naftali Tishby. Contact: Nati Srebro, TTI-C nati@tti-c.org 834-7493 -------------- next part -------------- An HTML attachment was scrubbed... URL: http://ttic.uchicago.edu/pipermail/colloquium/attachments/20090804/b0f3fea9/attachment-0001.htm From macglashan at tti-c.org Fri Aug 7 08:32:51 2009 From: macglashan at tti-c.org (Julia MacGlashan) Date: Fri Aug 7 08:33:01 2009 Subject: [TTIC Colloquium] Wonseok Dissertation Defense, 8-10-09 Message-ID: <37C6F64586B14738858516DD0AF06229@jmacglDPLFYD1> REMINDER: *** Toyota Technological Institute at Chicago *** *** Dissertation Defense *** Candidate: Wonseok Chae Date: Monday, August 10, 2009 Time and Location: 10:30 a.m. in TTI-C Conference Room #526 Title: Type Safe Extensible Programming Abstract: Software products evolve over time. Sometimes they evolve by adding new features, and sometimes by either fixing bugs or replacing outdated implementations with new ones. When software engineers fail to anticipate such evolution during development, they will eventually be forced to re-architect or re-build from scratch. Therefore, it has been common practice to prepare for changes so that software products are extensible over their lifetimes. However, making software extensible is challenging because it is difficult to anticipate successive changes and to provide adequate abstraction mechanisms over potential changes. Such extensibility mechanisms, furthermore, should not compromise any existing functionality during extension. Software engineers would benefit from a tool that provides a way to add extensions in a reliable way. It is natural to expect programming languages to serve this role. Extensible programming is one effort to address these issues. In my work, I present type safe extensible programming using the MLPolyR language. MLPolyR is an ML-like functional language whose type system provides type-safe extensibility mechanisms at several levels. After presenting the language, I will show how these extensibility mechanisms can be put to good use in the context of product line engineering. Product line engineering is an emerging software engineering paradigm that aims to manage variations, which originate from successive changes in software. Candidate's Advisor: Prof. Matthias Blume A draft copy of Wonseok Chae' dissertation will be available soon in TTI-C library From macglashan at tti-c.org Thu Aug 20 10:26:17 2009 From: macglashan at tti-c.org (Julia MacGlashan) Date: Thu Aug 20 10:26:24 2009 Subject: [TTIC Colloquium] TTI-C Colloquium: Nina Balcan (MSR New England & Georgia Tech) References: <9BFA4FFE1ACE407581765CA2326E1547@jmacglDPLFYD1> Message-ID: When: Monday, Aug 24 @ 4:00pm (NOTE: This is a special time) Where: 6045 S Kenwood Ave, TTI-C Conference Room #526 (5th Floor) Who: Maria Florina (Nina) Balcan (MSR New England and Georgia Tech) Title: The Dynamics of Equilibria Many natural games have both high and low cost Nash equilibria: their Price of Anarchy is high and yet their Price of Stability is low. In such cases, one could hope to "move" behavior from a high cost equilibrium to a low cost one by a "public service advertising campaign" encouraging players to follow the low-cost equilibrium. If every player follows the advice then we are done (since it's an equilibrium). However, the assumption that everyone follows instructions is unrealistic. A more natural assumption is that some players will follow them, while other players will not. In this paper we consider the question of to what extent can such an advertising campaign cause behavior to switch from a bad equilibrium to a good one even if only a fraction of people actually follow the given advice, and do so only temporarily. Unlike in the "price of altruism" model, we assume everyone will ultimately act in their own interest. We analyze this question for several important and widely studied classes of games including network design with fair cost sharing, scheduling with unrelated machines, and party affiliation games (which include consensus and cut games). We show that for some of these games (such as fair cost sharing), a random alpha fraction of the population following the given advice is sufficient to get a guarantee within any O(1/alpha) factor of the price of stability for any alpha > 0. However, for some games (such as party affiliation games), there is a strict threshold (in this case, alpha < 1/2 yields almost no benefit, yet alpha > 1/2 is enough to reach near-optimal behavior), and for some games, such as scheduling, no value alpha < 1 is sufficient. We conclude by analyzing an adaptive model in which rather than certain players following the advice and others behaving selfishly, the set of players following the given advice may change over time, as players test out the advice and make decisions on whether or not to follow it using adaptive learning behavior. This is based on work joint with Avrim Blum and Yishay Mansour. Contact: Nati Srebro, TTI-C nati@tti-c.org 834-7493 Speaker Schedule: http://www.tti-c.org/colloquium.php -------------- next part -------------- A non-text attachment was scrubbed... Name: winmail.dat Type: application/ms-tnef Size: 13570 bytes Desc: not available Url : http://ttic.uchicago.edu/pipermail/colloquium/attachments/20090820/9ad925d9/winmail.bin From macglashan at tti-c.org Mon Aug 24 08:09:18 2009 From: macglashan at tti-c.org (Julia MacGlashan) Date: Mon Aug 24 08:09:39 2009 Subject: [TTIC Colloquium] TTI-C Colloquium: Nina Balcan (MSR New England & Georgia Tech) References: <9BFA4FFE1ACE407581765CA2326E1547@jmacglDPLFYD1> Message-ID: <3BA1051F764143879DD463F7455EF08F@jmacglDPLFYD1> REMINDER When: Monday, Aug 24 @ 4:00pm (NOTE: This is a special time) Where: 6045 S Kenwood Ave, TTI-C Conference Room #526 (5th Floor) Who: Maria Florina (Nina) Balcan (MSR New England and Georgia Tech) Title: The Dynamics of Equilibria Many natural games have both high and low cost Nash equilibria: their Price of Anarchy is high and yet their Price of Stability is low. In such cases, one could hope to "move" behavior from a high cost equilibrium to a low cost one by a "public service advertising campaign" encouraging players to follow the low-cost equilibrium. If every player follows the advice then we are done (since it's an equilibrium). However, the assumption that everyone follows instructions is unrealistic. A more natural assumption is that some players will follow them, while other players will not. In this paper we consider the question of to what extent can such an advertising campaign cause behavior to switch from a bad equilibrium to a good one even if only a fraction of people actually follow the given advice, and do so only temporarily. Unlike in the "price of altruism" model, we assume everyone will ultimately act in their own interest. We analyze this question for several important and widely studied classes of games including network design with fair cost sharing, scheduling with unrelated machines, and party affiliation games (which include consensus and cut games). We show that for some of these games (such as fair cost sharing), a random alpha fraction of the population following the given advice is sufficient to get a guarantee within any O(1/alpha) factor of the price of stability for any alpha > 0. However, for some games (such as party affiliation games), there is a strict threshold (in this case, alpha < 1/2 yields almost no benefit, yet alpha > 1/2 is enough to reach near-optimal behavior), and for some games, such as scheduling, no value alpha < 1 is sufficient. We conclude by analyzing an adaptive model in which rather than certain players following the advice and others behaving selfishly, the set of players following the given advice may change over time, as players test out the advice and make decisions on whether or not to follow it using adaptive learning behavior. This is based on work joint with Avrim Blum and Yishay Mansour. Contact: Nati Srebro, TTI-C nati@tti-c.org 834-7493 Speaker Schedule: http://www.tti-c.org/colloquium.php -------------- next part -------------- A non-text attachment was scrubbed... Name: winmail.dat Type: application/ms-tnef Size: 13562 bytes Desc: not available Url : http://ttic.uchicago.edu/pipermail/colloquium/attachments/20090824/2a65a92e/winmail-0001.bin