From macglashan at tti-c.org Mon Feb 2 11:07:44 2009 From: macglashan at tti-c.org (Julia MacGlashan) Date: Mon Feb 2 11:06:52 2009 Subject: [TTIC Colloquium] TTI-C Talk: Xi Chen (Princeton University) Message-ID: *** TTI-C moved! Come see our new building: http://maps.uchicago.edu/farsoutheast/ *** When: Tuesday, February 3 @ 11:00am (lunch will be provided after talk) Where: TTI-C Conference Room #526: 6045 S Kenwood Ave, 5th Floor Who: Xi Chen (Princeton University) Title: On the Computation and Approximation of Two-Player Nash Equilibria In 1950, Nash showed that every non-cooperative game has an equilibrium. Before his work, the result was known only for two-player zero-sum games. While von Neumann's minimax theorem was the mathematical foundation of the two-player zero-sum game, Brouwer's and Katutani's fixed point theorems were crucially used in Nash's proofs. His approach was later used by Arrow and Debreu in their development of the General Equilibrium Theory. Fourteen years after Nash's work, Lemke and Howson designed a simplex-like path-following algorithm for finding a Nash equilibrium in a general two-player game. Despite its several appealing properties, this algorithm was recently shown to have exponential worst-case complexity. In contrast, in his groundbreaking work, Khachiyan showed that the ellipsoid algorithm can solve any linear program and hence any two-player zero-sum game in polynomial time. However, this success has not been extended to general two-player games --- no polynomial-time algorithm has yet been found for this remarkable problem. In this talk, I will present results that characterize the complexity of computing and approximating two-player Nash equilibria (Joint work with Xiaotie Deng and Shang-Hua Teng). Contact: Lance Fortnow fortnow@eecs.northwestern.edu -------------- next part -------------- An HTML attachment was scrubbed... URL: http://ttic.uchicago.edu/pipermail/colloquium/attachments/20090202/6e95a0ca/attachment.htm From macglashan at tti-c.org Tue Feb 3 08:57:20 2009 From: macglashan at tti-c.org (Julia MacGlashan) Date: Tue Feb 3 08:56:23 2009 Subject: [TTIC Colloquium] TTI-C Talk: Prasad Raghavendra (University of Washington) References: <9BFA4FFE1ACE407581765CA2326E1547@jmacglDPLFYD1> Message-ID: <46DC4CA5A2A0495DAFEEE135D184FB03@jmacglDPLFYD1> *** TTI-C moved! Come see our new building: http://maps.uchicago.edu/farsoutheast/ *** When: Wednesday, February 4 @ 11:00am (lunch will be provided after talk) Where: TTI-C Conference Room #526, 6045 S Kenwood Ave, 5th Floor Who: Prasad Raghavendra (University of Washington) Title: Approximating the optimum: Efficient algorithms and their limits Most combinatorial optimization problems of interest are NP-hard to solve exactly. To cope with this intractability, one settles for approximation algorithms with provable guarantee on the quality of approximation. Despite great success in designing approximation algorithms, underlying a vast majority of the work is the technique of linear programming, or more generally semi-definite programming. This poses the natural question: How far can we push the technique of semi-definite programming? Are there techniques that yield better approximations than semi-definite programming? In this work, we show that the simplest semi-definite programs yield the best approximation for large classes of optimization problems ranging from constraint satisfaction problems (CSP) to graph labeling problems, ordering CSPs to certain clustering problems, under the Unique Games Conjecture. In a surprising convergence of algorithms and hardness results, the same work also leads to a generic algorithm that achieves the optimal approximation for every constraint satisfaction problem. Bio: Prasad Raghavendra obtained his bachelor's degree at Indian Institute of Technology, Madras, and is currently a Phd candidate at University of Washington. His research spans several areas of theoretical computer science including approximation algorithms and inapproximability results for NP-hard optimization problems, error correcting codes, metric embeddings, computational learning and complexity. He is the recipient of the Best Paper and Best Student Paper awards at STOC 08. Contact: Julia Chuzhoy, TTI-C cjulia@tti-c.org 834-2490 -------------- next part -------------- A non-text attachment was scrubbed... Name: winmail.dat Type: application/ms-tnef Size: 8290 bytes Desc: not available Url : http://ttic.uchicago.edu/pipermail/colloquium/attachments/20090203/9a0de375/winmail.bin From macglashan at tti-c.org Wed Feb 4 10:02:37 2009 From: macglashan at tti-c.org (Julia MacGlashan) Date: Wed Feb 4 10:01:35 2009 Subject: [TTIC Colloquium] TTI-C Talk: Yury Makarychev (Microsoft Research) References: <9BFA4FFE1ACE407581765CA2326E1547@jmacglDPLFYD1> Message-ID: *** TTI-C moved! Come see our new building: http://maps.uchicago.edu/farsoutheast/ *** When: Thursday, February 5 @ 11:00am (lunch will be provided after talk) Where: TTI-C Conference Room #526, 6045 S Kenwood Ave, 5th Floor Who: Yury Makarychev (Microsoft Research) Title: Approximation Algorithms, Semidefinite Programming and Unique Games Best known approximation algorithms for many combinatorial optimization problems are based on semidefinite (SDP) programming. Are these algorithms optimal? Are there any techniques better than SDP? We do not know the answer. However, the Unique Games Conjecture (UGC) implies that for a wide class of problems known SDP algorithms are indeed optimal. One of the central questions in the theory of approximation algorithms is whether this conjecture is true or false. To answer this question we need to better understand Unique Games and design good approximation algorithms for them. In the first part of the talk, we will describe approximation algorithms for Unique Games that significantly improve previously known approximation guarantees and that are asymptotically optimal assuming UGC. Then we will discuss other possible approaches to Unique Games. We will describe the Sherali-Adams (SA) hierarchy of relaxations for combinatorial optimization problems (the strongest LP hierarchy considered in the literature). We will prove negative results for MAX CUT, Vertex Cover, Max Acyclic Subgraph and other problems. In particular, we will show that algorithms based on SA relaxations cannot disprove UGC. Based on joint papers with Moses Charikar, Eden Chlamtac, and Konstantin Makarychev. Contact: Julia Chuzhoy, TTI-C cjulia@tti-c.org 834-2490 -------------- next part -------------- A non-text attachment was scrubbed... Name: winmail.dat Type: application/ms-tnef Size: 8130 bytes Desc: not available Url : http://ttic.uchicago.edu/pipermail/colloquium/attachments/20090204/28194fab/winmail-0001.bin From macglashan at tti-c.org Thu Feb 12 13:51:06 2009 From: macglashan at tti-c.org (Julia MacGlashan) Date: Thu Feb 12 13:49:43 2009 Subject: [TTIC Colloquium] TTI-C Talk: Rishi Saket (Georgia Tech) References: <9BFA4FFE1ACE407581765CA2326E1547@jmacglDPLFYD1> Message-ID: <5CC36A9E47A8437BB891870DE8C7E4DB@jmacglDPLFYD1> When: Thursday, February 19 @ 11:00am (lunch will be provided after talk) Where: TTI-C Conference Room #526, 6045 S Kenwood Ave, 5th Floor Who: Rishi Saket (Georgia Tech) Title: Hardness of learning common concept classes We study the inapproximability of accurately learning some common concept classes in the Probably Approximately Correct (PAC) model. In particular we show the following results: 1. Intersection of two halfspaces is hard to PAC-learn beyond an accuracy of 1/2 + \eps by an intersection of constantly many halfspaces. This improves the previous NP-hardness result [ABFKP04] to an optimal inapproximability factor. 2. Noisy parity (linear function) over GF(2) is hard to PAC-learn beyond an accuracy of 1 - 2^{-d} + \eps by a degree d polynomial. This extends the previous inapproximability for learning noisy parity by a parity given by Hastad's 3-bit PCP [Hastad01]. 3. A two term DNF is hard to PAC-learn beyond an accuracy of 1/2 + \eps by a constant term DNF. This improves the previous NP-hardness result [ABFKP04] to an optimal inapproximability factor. We also show the same inapproximability for learning a noisy AND by a CNF with constant clause size which extends and simplifies previous results [FGKP06]. 4. Given the truth table of a boolean function f over d variables, it is hard to find an equivalent DNF formula for f with minimum number of terms to within d^{1-\eps} of the optimal, for any \eps > 0. This improves the previous best d^{\gamma} (\gamma bounded away from 1) inapproximability [Feldman06][AHMPS06] to an essentially optimal factor as the greedy Set-Cover algorithm gives an O(d) approximation. The talk will define the PAC-learning framework, give an overview of the techniques used for proving these results and sketch the reductions for some of them. This is based on joint works with Subhash Khot and Parikshit Gopalan. Contact: Julia Chuzhoy, TTI-C cjulia@tti-c.org 834-2490 -------------- next part -------------- A non-text attachment was scrubbed... Name: winmail.dat Type: application/ms-tnef Size: 8058 bytes Desc: not available Url : http://ttic.uchicago.edu/pipermail/colloquium/attachments/20090212/c2b20d72/winmail.bin From macglashan at tti-c.org Tue Feb 17 11:09:40 2009 From: macglashan at tti-c.org (Julia MacGlashan) Date: Tue Feb 17 10:55:24 2009 Subject: [TTIC Colloquium] TTI-C Talk: Alexandr Andoni (MIT) References: <9BFA4FFE1ACE407581765CA2326E1547@jmacglDPLFYD1> Message-ID: <35830396774E4A55B2A2008CFF822092@jmacglDPLFYD1> When: TOMORROW: Wednesday, February 18 @ 11:00am (lunch will be provided after talk) Where: TTI-C Conference Room #526, 6045 S Kenwood Ave, 5th Floor Who: Alexandr Andoni (MIT) Title: Nearest Neighbor Search: Old and New Approaches When working with massive datasets, a frequently recurring problem is that of Nearest Neighbor Search (NNS) in high-dimensional spaces. In this problem, the goal is to preprocess a set of objects (e.g., images), so that later, given a new query object, one can efficiently return the object most similar to the query. This problem is of significant importance in several areas such as machine learning, information retrieval, image/video/music clustering, code clone detection, etc. We develop new algorithms for NNS under several important distances, via both old, tested paradigms as well as newly developed paradigms. In the first part, we give a new near-optimal algorithm for NNS under the classical Euclidean distance. This algorithm is based on Locality-Sensitive Hashing, a scheme that has already been successfully used in a number of practical scenarios. Our algorithm is near-optimal in the class of such hashing algorithms. In the second part, we propose a new approach to designing NNS algorithms, for the scenarios where the above hashing approach is provably impossible. Our approach has already been applied for designing NNS under a variant of edit distance, yielding an exponential improvement in approximation over what is even possible via the old approaches. Furthermore, the key idea, embedding into product metrics, has also been used for other applications, such as approximating the edit distance between two strings in near-linear time. Contact: Julia Chuzhoy, TTI-C cjulia@tti-c.org 834-2490 -------------- next part -------------- A non-text attachment was scrubbed... Name: winmail.dat Type: application/ms-tnef Size: 8078 bytes Desc: not available Url : http://ttic.uchicago.edu/pipermail/colloquium/attachments/20090217/ff1a3259/winmail.bin From macglashan at tti-c.org Wed Feb 18 09:25:35 2009 From: macglashan at tti-c.org (Julia MacGlashan) Date: Wed Feb 18 09:11:13 2009 Subject: [TTIC Colloquium] TTI-C Talk: Rishi Saket (Georgia Tech) References: <9BFA4FFE1ACE407581765CA2326E1547@jmacglDPLFYD1> Message-ID: When: TOMORROW: Thursday, February 19 @ 11:00am (lunch will be provided after talk) Where: TTI-C Conference Room #526, 6045 S Kenwood Ave, 5th Floor Who: Rishi Saket (Georgia Tech) Title: Hardness of learning common concept classes We study the inapproximability of accurately learning some common concept classes in the Probably Approximately Correct (PAC) model. In particular we show the following results: 1. Intersection of two halfspaces is hard to PAC-learn beyond an accuracy of 1/2 + \eps by an intersection of constantly many halfspaces. This improves the previous NP-hardness result [ABFKP04] to an optimal inapproximability factor. 2. Noisy parity (linear function) over GF(2) is hard to PAC-learn beyond an accuracy of 1 - 2^{-d} + \eps by a degree d polynomial. This extends the previous inapproximability for learning noisy parity by a parity given by Hastad's 3-bit PCP [Hastad01]. 3. A two term DNF is hard to PAC-learn beyond an accuracy of 1/2 + \eps by a constant term DNF. This improves the previous NP-hardness result [ABFKP04] to an optimal inapproximability factor. We also show the same inapproximability for learning a noisy AND by a CNF with constant clause size which extends and simplifies previous results [FGKP06]. 4. Given the truth table of a boolean function f over d variables, it is hard to find an equivalent DNF formula for f with minimum number of terms to within d^{1-\eps} of the optimal, for any \eps > 0. This improves the previous best d^{\gamma} (\gamma bounded away from 1) inapproximability [Feldman06][AHMPS06] to an essentially optimal factor as the greedy Set-Cover algorithm gives an O(d) approximation. The talk will define the PAC-learning framework, give an overview of the techniques used for proving these results and sketch the reductions for some of them. This is based on joint works with Subhash Khot and Parikshit Gopalan. Contact: Julia Chuzhoy, TTI-C cjulia@tti-c.org 834-2490 -------------- next part -------------- A non-text attachment was scrubbed... Name: winmail.dat Type: application/ms-tnef Size: 8022 bytes Desc: not available Url : http://ttic.uchicago.edu/pipermail/colloquium/attachments/20090218/f7e3df1f/winmail-0001.bin From macglashan at tti-c.org Fri Feb 20 15:50:20 2009 From: macglashan at tti-c.org (Julia MacGlashan) Date: Fri Feb 20 15:35:53 2009 Subject: [TTIC Colloquium] TTI-C Talk: Jay Bardhan (Argonne National Lab) References: <9BFA4FFE1ACE407581765CA2326E1547@jmacglDPLFYD1> Message-ID: ***TTI-C Moved! This talk will be held at our new address. You can view a map here: http://maps.uchicago.edu/farsoutheast/*** When: Wednesday, February 25 @ 11:00am (lunch will be provided after talk) Where: TTI-C Conference Room #526, 6045 S Kenwood Ave, 5th Floor Who: Jaydeep Bardhan Talk (Argonne National Lab) Title: Numerical Algorithms in Molecular Science and Engineering Without question, the advent of high-speed computers has revolutionized biology, allowing both the modeling of complex phenomena and the analysis of high-dimensional, high-resolution experimental data. In this talk, I will describe my research in an area that often receives less attention than the computational results themselves--the numerical algorithms used in biophysical modeling. Examples from recent work illustrate how the development of numerical algorithms can offer new insights into underlying physics and generate important new directions for research by reshaping the landscape of computational investigation. First, I have been working to improve methods for modeling electrostatic interactions within and between molecules in aqueous solution. Most recently, I have proved previously unrecognized equivalences between different approaches to the problem. In addition, my recent demonstration of the intimate relationship between boundary-element simulations and popular Generalized-Born (GB) models may allow significant improvements in accuracy without sacrificing computational efficiency. A second project builds on these methods for electrostatic analysis, in an effort to develop improved strategies for molecular design. Our novel PDE-constrained approach to this optimization dramatically reduces the computational time required to solve problems, opening up new research opportunities that were previously infeasible. Taken together, these efforts demonstrate the important and enabling roles that numerical algorithms can play in the development of molecular science and engineering. Contact: Jinbo Xu, TTI-C j3xu@tti-c.org 834-2511 -------------- next part -------------- An HTML attachment was scrubbed... URL: http://ttic.uchicago.edu/pipermail/colloquium/attachments/20090220/ff5ce787/attachment.htm From cnovak at tti-c.org Mon Feb 23 12:30:10 2009 From: cnovak at tti-c.org (Christina Novak) Date: Mon Feb 23 12:30:47 2009 Subject: [TTIC Colloquium] TTI-C Distinguished Lecture Series (March 3, Dr. Mihalis Yannakakis- Columbia University) Message-ID: <5824D1D232704FF39B9670F277D35E5B@cnovakHBRQFD1> Good afternoon, Please join us for Toyota Technological Institute at Chicago's second 2009 Distinguished Lecture Series speaker, Dr. Mihalis Yannakakis. The talk will be from 2 pm, and located in TTI-C's new facility at 6045 S. Kenwood Ave. More details, speaker bios and directions may be found at http://tti-c.org/dls. We look forward to seeing you at the lecture! Sincerely, David McAllester TTI-C Chief Academic Officer For questions about the TTI-C Distinguished Lecture Series, please contact: Chrissy Novak cnovak@tti-c.org or (773)834-2216. Tuesday, March 3rd, 2009 (2 pm) Mihalis Yannakakis (Columbia University) "Equilibria, Fixed Points, and Complexity Classes" Abstract: Many models from a variety of areas involve the computation of an equilibrium or fixed point of some kind. Examples include Nash equilibria in games; market equilibria; computing optimal strategies and the values of competitive games (stochastic and other games); stable configurations of neural networks; analyzing basic stochastic models for evolution like branching processes and for language like stochastic context-free grammars; and models that incorporate the basic primitives of probability and recursion like recursive Markov chains. It is not known whether these problems can be solved in polynomial time. Despite their broad diversity, there are certain common computational principles that underlie different types of equilibria and connect many of these problems to each other. In this talk we will discuss these common principles and the corresponding complexity classes that capture them. -------------- next part -------------- An HTML attachment was scrubbed... URL: http://ttic.uchicago.edu/pipermail/colloquium/attachments/20090223/63bb756f/attachment.htm From macglashan at tti-c.org Mon Feb 23 13:57:07 2009 From: macglashan at tti-c.org (Julia MacGlashan) Date: Mon Feb 23 13:42:30 2009 Subject: [TTIC Colloquium] TTI-C Colloquium: Shmuel Friedland, UIC References: <9BFA4FFE1ACE407581765CA2326E1547@jmacglDPLFYD1> Message-ID: <0BD4273E57B849C48169BB98F436F5D3@jmacglDPLFYD1> When: Monday, March 2 @ 2:00pm Where: TTI-C Conference Room #526, 6045 S Kenwood Ave, 5th Floor Who: Shmuel Friedland, UIC Title: Tensors: theory and applications Since the pioneering work of Tucker in 60's on 3-mode factor analysis there is a tremendous rise in popularity of using tensors in engineering applications: factor analysis, video tracking; computer science: data analysis, face recognition; numerical mathematics: fast multiplication of matrices, low rank approximations; and mathematics: secant varieties, Perron-Frobenius theory. In this talk I will discuss some basic notions in tensors, mention some results, applications and open problems mainly for 3-mode tensors. Contact: Shai Shalev-Shwartz, TTI-C shai@tti-c.org 834-6850 -------------- next part -------------- A non-text attachment was scrubbed... Name: winmail.dat Type: application/ms-tnef Size: 7618 bytes Desc: not available Url : http://ttic.uchicago.edu/pipermail/colloquium/attachments/20090223/368256be/winmail-0001.bin From macglashan at tti-c.org Thu Feb 26 13:50:28 2009 From: macglashan at tti-c.org (Julia MacGlashan) Date: Thu Feb 26 13:35:44 2009 Subject: [TTIC Colloquium] TTI-C Talk: Tamir Hazan (Hebrew University of Jerusalem) References: <9BFA4FFE1ACE407581765CA2326E1547@jmacglDPLFYD1> Message-ID: <5B4F2A0E836742E7A992564C111A2B48@jmacglDPLFYD1> When: Wednesday, March 4 @ 11:00am (lunch will be provided after talk) Where: TTI-C Conference Room #526, 6045 S Kenwood Ave, 5th Floor Who: Tamir Hazan (Hebrew University of Jerusalem) Title: Convex Belief Propagation - Approximated inference and LP-relaxations We derive a one-parameter local message-passing algorithm, called "norm-product", which covers both the tasks of computing approximate marginal probabilities and maximum a posteriori (MAP) assignment for general graphical models. A parameter $\epsilon$ controls a perturbation term of a "fractional entropy approximation" $\tilde H$ which includes Bethe, Tree-reweighted (TRW) and convex entropy approximations. When $\tilde H$ is the Bethe approximation, the settings $\epsilon=0$ and $\epsilon=1$ produce the max-product and sum-product algorithms, respectively. When $\tilde H$ is a convex entropy approximation and $\epsilon\rightarrow 0$, the algorithm is a globally convergent Linear Programming (LP) relaxation of the MAP problem. When $\tilde H$ is convex and $\epsilon=1$, norm-product is a globally convergent algorithm for "convex free energies" for approximate marginal probabilities, and when $\epsilon=0$ norm-product becomes a family of convergent "max-product-like" algorithms for computing approximate MAP. Contact: Shai Shalev-Shwartz, TTI-C shai@tti-c.org 834-6850 -------------- next part -------------- A non-text attachment was scrubbed... Name: winmail.dat Type: application/ms-tnef Size: 7946 bytes Desc: not available Url : http://ttic.uchicago.edu/pipermail/colloquium/attachments/20090226/73862a36/winmail.bin From macglashan at tti-c.org Fri Feb 27 09:08:08 2009 From: macglashan at tti-c.org (Julia MacGlashan) Date: Fri Feb 27 08:53:19 2009 Subject: [TTIC Colloquium] TTI-C Talk: Satyen Kale (Microsoft) References: <9BFA4FFE1ACE407581765CA2326E1547@jmacglDPLFYD1> Message-ID: <57703BC3B66E4964BBFA8660E0DD3C67@jmacglDPLFYD1> When: Thursday, March 5 @ 11:00am (lunch will be provided after talk) Where: TTI-C Conference Room #526, 6045 S Kenwood Ave, 5th Floor Who: Satyen Kale (Microsoft Research New England) Title: Improved Decision-Making under Uncertainty Decision-making in the face of uncertainty over future outcomes is a fundamental algorithmic task, with roots in statistics and information theory, and applications in machine learning, signal processing, network routing and finance. The framework of regret minimization captures the notion of online decision-making algorithms that are competitive with the best possible decision in hindsight, under minimal assumptions on how the costs of the decisions are set. A major achievement of online learning theory has been the development of algorithms that minimize regret even under such "worst-case" assumptions. However, the regret bounds are quite suboptimal in real-life scenarios where the decision costs are particularly benign. On the other hand, "average-case" learning methods that posit a specific stochastic model on the costs have better convergence bounds, but may fail to work when the actual costs deviate from the model. Although these algorithms have been developed over several decades, designing regret minimizing algorithms that smoothly handle benign cost sequences, while not compromising worst-case robustness, was considered a significant open problem. In my talk, I will give an overview of my recent work with Elad Hazan which solved this open problem for four fundamental online learning scenarios: (a) prediction from expert advice, (b) online linear optimization, (c) universal portfolio selection, and (d) bandit linear optimization. The work on portfolio selection algorithms has implications in the standard Brownian motion model of stock prices, which were verified by experiments on real data. Contact: Sham Kakade, TTI-C sham@tti-c.org 834-2550 -------------- next part -------------- A non-text attachment was scrubbed... Name: winmail.dat Type: application/ms-tnef Size: 8390 bytes Desc: not available Url : http://ttic.uchicago.edu/pipermail/colloquium/attachments/20090227/ecd26c84/winmail.bin From macglashan at tti-c.org Fri Feb 27 09:26:42 2009 From: macglashan at tti-c.org (Julia MacGlashan) Date: Fri Feb 27 09:11:48 2009 Subject: [TTIC Colloquium] TTI-C Colloquium: Shmuel Friedland, UIC References: <9BFA4FFE1ACE407581765CA2326E1547@jmacglDPLFYD1> Message-ID: REMINDER When: Monday, March 2 @ 2:00pm Where: TTI-C Conference Room #526, 6045 S Kenwood Ave, 5th Floor Who: Shmuel Friedland, UIC Title: Tensors: theory and applications Since the pioneering work of Tucker in 60's on 3-mode factor analysis there is a tremendous rise in popularity of using tensors in engineering applications: factor analysis, video tracking; computer science: data analysis, face recognition; numerical mathematics: fast multiplication of matrices, low rank approximations; and mathematics: secant varieties, Perron-Frobenius theory. In this talk I will discuss some basic notions in tensors, mention some results, applications and open problems mainly for 3-mode tensors. Contact: Shai Shalev-Shwartz, TTI-C shai@tti-c.org 834-6850 -------------- next part -------------- A non-text attachment was scrubbed... Name: winmail.dat Type: application/ms-tnef Size: 7506 bytes Desc: not available Url : http://ttic.uchicago.edu/pipermail/colloquium/attachments/20090227/64ecaafd/winmail-0001.bin