From pondabarnes at tti-c.org Tue May 1 09:26:04 2007 From: pondabarnes at tti-c.org (Ponda Barnes) Date: Tue May 1 09:30:00 2007 Subject: [TTIC Colloquium] FW: Guest Speaker Announcement Message-ID: <001a01c78bfc$a6c3ae90$e8bf8780@TTIC47> Reminder!! Guest Speaker Presented by: Toyota Technological Institute at Chicago Speaker: Sorin Draghici Speaker's home page: http://vortex.cs.wayne.edu/sorin/index.htm Date: Tuesday, May 1, 2007 Time: 10:00 Location: TTI-C Conference room Title: A systems biology approach to pathway analysis Abstract: A common challenge in the analysis of genomics data is trying to understand the underlying phenomenon in the context of all complex interactions taking place on various signaling pathways. A statistical approach using various models is universally used to identify the most relevant pathways in a given experiment. In this talk, we show that despite its general adoption, this statistical analysis is unsatisfactory, and can often provide incorrect results. We discuss how and why the limited numerical nature of this approach makes it ill suited to cope with the complex interactions and dependencies that characterize living organisms. Using a systems biology approach, we developed a more powerful impact analysis that extends the classical statistical approach by incorporating a number of crucial biological factors such as the magnitude of the expression change for each gene, the type and the position of the genes in the given pathways, the interactions between them, etc. To the best of our knowledge, none of the other approaches currently used for pathway analysis is able to integrate these factors in a coherent model. Notably, the novel integrated model we propose is fully coherent with the classical approach. When the limitations of the classical approach are forcefully imposed (e.g., ignoring the magnitude of the measured expression changes or ignoring the regulatory interactions between genes), the impact analysis reduces to the classical approach and yields exactly the same results. On several illustrative data sets, the classical analysis produces both false positives and false negatives while the impact analysis provides biologically meaningful results. If you have any questions or would like to meet the speaker, please contact Ponda Barnes at pondabarnes@tti-c.org For future TTI-C talks and events please go to http://ttic.uchicago.edu/cal/month.php -------------- next part -------------- An HTML attachment was scrubbed... URL: http://ttic.uchicago.edu/pipermail/colloquium/attachments/20070501/b589aa0d/attachment.htm From pondabarnes at tti-c.org Thu May 3 09:20:48 2007 From: pondabarnes at tti-c.org (Ponda Barnes) Date: Thu May 3 09:26:09 2007 Subject: [TTIC Colloquium] Guest Speaker Announcement Message-ID: <002401c78d8e$3f803df0$e8bf8780@TTIC47> Guest Speaker Presented by: Toyota Technological Institute at Chicago Speaker: Ali Jadbabaie Speaker's homepage: http://www.seas.upenn.edu/~jadbabai/ Date: Thursday, May 03, 2007 Time: 3:00 Location: TTI-C Conference room Title: Distributed Coordination: From Flocking and Synchronization to Coverage in Sensor Networks Abstract: In this talk, we provide a unified view of several distributed coordination and consensus algorithms, which have appeared in various disciplines such as distributed systems, statistical physics, biology, computer graphics, robotics, and control theory over the past 2 decades. These algorithms have been proposed as a mechanism for demonstrating emergence of a global collective behavior (such as social aggregation in animals, schooling, flocking and synchronization in oscillator networks) using purely local interactions. Utilizing tools from spectral graph theory and control and dynamical systems theory, we provide an analysis of these algorithms. Using tools from algebraic topology, we extend our results from graphs to simplicial complexes to verify coverage in mobile sensor networks in a decentralized fashion. These simplicial complexes are induced by the local connectivity of agents in a network, and their homology groups allow us to infer the coverage properties of mobile sensort networks with time-varying interconnections. The enabling mathematical technique for our result is the theory of higher order Laplacian operators, which will be presented as a generalization of the graph Laplacian used in the first part of the talk for analysis of synchronization, agreement and consensus problems. If you have any questions or would like to meet the speaker, please contact Ponda Barnes at pondabarnes@tti-c.org For future TTI-C talks and events, please go to http://ttic.uchicago.edu/cal/month.php -------------- next part -------------- An HTML attachment was scrubbed... URL: http://ttic.uchicago.edu/pipermail/colloquium/attachments/20070503/84b1e877/attachment-0001.htm From pondabarnes at tti-c.org Thu May 3 14:53:47 2007 From: pondabarnes at tti-c.org (Ponda Barnes) Date: Thu May 3 14:58:58 2007 Subject: [TTIC Colloquium] Guest Speaker announcement Message-ID: <005901c78dbc$c5919910$e8bf8780@TTIC47> Guest Speaker Presented by: Toyota Technological Institute at Chicago Speaker: Andrej Bogdanov Speaker's home page: http://dimacs.rutgers.edu/~adib/ Date: Friday, May 4, 2007 Time: 10:00 Location: TTI-C Conference room Title: Average-case hardness for NP Abstract: The gold standard of difficulty for a computational problem is NP hardness. However, NP hard instances of problems are quite atypical. In practice, and in many theoretical settings as well, one sometimes adopts the view that NP hard instances are so rare that they present no serious obstacle to the design of efficient algorithms Yet experience tells us that some NP problems are intractable even on typical instances. This assumption is a fundamental axiom of modern cryptography. So far computer science has been unable to explain why this happens. In this talk, I will focus on two basic questions: * What is a typical instance? * Can the extensive theory of NP hardness be used to explain why for certain problems typical instances are also hard? These two questions will give a glimpse of the central role that average-case complexity plays in the theory of computation. If you have any questions or would like to meet the speaker, please contact Ponda Barnes at pondabarnes@tti-c.org For future TTI-C talks and events please got to http://ttic.uchicago.edu/cal/month.php -------------- next part -------------- An HTML attachment was scrubbed... URL: http://ttic.uchicago.edu/pipermail/colloquium/attachments/20070503/804ed093/attachment.htm From pondabarnes at tti-c.org Fri May 4 09:26:05 2007 From: pondabarnes at tti-c.org (Ponda Barnes) Date: Fri May 4 09:31:53 2007 Subject: [TTIC Colloquium] FW: Guest Speaker announcement Message-ID: <001b01c78e58$270e66f0$e8bf8780@TTIC47> Reminder!! Guest Speaker Presented by: Toyota Technological Institute at Chicago Speaker: Andrej Bogdanov Speaker's home page: http://dimacs.rutgers.edu/~adib/ Date: Friday, May 4, 2007 Time: 10:00 Location: TTI-C Conference room Title: Average-case hardness for NP Abstract: The gold standard of difficulty for a computational problem is NP hardness. However, NP hard instances of problems are quite atypical. In practice, and in many theoretical settings as well, one sometimes adopts the view that NP hard instances are so rare that they present no serious obstacle to the design of efficient algorithms Yet experience tells us that some NP problems are intractable even on typical instances. This assumption is a fundamental axiom of modern cryptography. So far computer science has been unable to explain why this happens. In this talk, I will focus on two basic questions: * What is a typical instance? * Can the extensive theory of NP hardness be used to explain why for certain problems typical instances are also hard? These two questions will give a glimpse of the central role that average-case complexity plays in the theory of computation. If you have any questions or would like to meet the speaker, please contact Ponda Barnes at pondabarnes@tti-c.org For future TTI-C talks and events please got to http://ttic.uchicago.edu/cal/month.php -------------- next part -------------- An HTML attachment was scrubbed... URL: http://ttic.uchicago.edu/pipermail/colloquium/attachments/20070504/e01b92df/attachment.htm From pondabarnes at tti-c.org Fri May 4 15:01:48 2007 From: pondabarnes at tti-c.org (Ponda Barnes) Date: Fri May 4 15:07:48 2007 Subject: [TTIC Colloquium] Guest Speaker Announcement Message-ID: <004701c78e87$0ee58340$e8bf8780@TTIC47> Guest Speaker Presented by: Toyota Technological Institute at Chicago Speaker: Nayantara Bhatnagar Speaker home page: http://www-static.cc.gatech.edu/~nand/ Date: 5/7/07 Time: 10:00 Location: TTI-C Conference room Title: Enhancing the Markov Chain Monte Carlo Method. Abstract: The Markov Chain Monte Carlo method is arguably the most powerful algorithmic tool available for approximate counting problems. Most known algorithms for such problems follow the paradigm of defining a Markov chain and showing that it mixes rapidly. However, there are natural counting problems where the obvious Markov chains do not mix rapidly. Annealing and Simulated Tempering are two heuristic approaches that can be applied in such situations. Both aim at finding ways to circumvent bottlenecks that cause Markov chains to mix slowly. In this talk, we will explore the power and limitations of these approaches. We present a simulated annealing based algorithm for the problem of generating random binary contingency tables. This problem can be restated as generating random bipartite graphs with a given degree sequence. This is based on joint work with Ivona Bezakova and Eric Vigoda. On the flip side, we show that in some scenarios, simulated tempering fails to speed up the mixing of the Markov chain and in fact no temperature based interpolants for the tempering algorithm can succeed. This is based on joint work with Dana Randall. If you have any questions or would like to meet the speaker, please contact Ponda Barnes at pondabarnes@tti-c. -------------- next part -------------- An HTML attachment was scrubbed... URL: http://ttic.uchicago.edu/pipermail/colloquium/attachments/20070504/1a3618da/attachment-0001.htm From pondabarnes at tti-c.org Mon May 7 09:01:18 2007 From: pondabarnes at tti-c.org (Ponda Barnes) Date: Mon May 7 09:08:56 2007 Subject: [TTIC Colloquium] FW: Guest Speaker Announcement Message-ID: <001b01c790b0$2f78acb0$e8bf8780@TTIC47> Reminder!! Guest Speaker Presented by: Toyota Technological Institute at Chicago Speaker: Nayantara Bhatnagar Speaker home page: http://www-static.cc.gatech.edu/~nand/ Date: 5/7/07 Time: 10:00 Location: TTI-C Conference room Title: Enhancing the Markov Chain Monte Carlo Method. Abstract: The Markov Chain Monte Carlo method is arguably the most powerful algorithmic tool available for approximate counting problems. Most known algorithms for such problems follow the paradigm of defining a Markov chain and showing that it mixes rapidly. However, there are natural counting problems where the obvious Markov chains do not mix rapidly. Annealing and Simulated Tempering are two heuristic approaches that can be applied in such situations. Both aim at finding ways to circumvent bottlenecks that cause Markov chains to mix slowly. In this talk, we will explore the power and limitations of these approaches. We present a simulated annealing based algorithm for the problem of generating random binary contingency tables. This problem can be restated as generating random bipartite graphs with a given degree sequence. This is based on joint work with Ivona Bezakova and Eric Vigoda. On the flip side, we show that in some scenarios, simulated tempering fails to speed up the mixing of the Markov chain and in fact no temperature based interpolants for the tempering algorithm can succeed. This is based on joint work with Dana Randall. If you have any questions or would like to meet the speaker, please contact Ponda Barnes at pondabarnes@tti-c.org For future TTI-C talks and events, please visit http://ttic.uchicago.edu/cal/month.php -------------- next part -------------- An HTML attachment was scrubbed... URL: http://ttic.uchicago.edu/pipermail/colloquium/attachments/20070507/8351d541/attachment.htm From pondabarnes at tti-c.org Tue May 8 08:23:41 2007 From: pondabarnes at tti-c.org (Ponda Barnes) Date: Tue May 8 08:31:54 2007 Subject: [TTIC Colloquium] Guest Speaker Announcement Message-ID: <000901c79174$18ac5710$e8bf8780@TTIC47> Guest Speaker Presented by: Toyota Technological Institute at Chicago Speaker: Hubert Chan Speaker's home page: http://www.cs.uwaterloo.ca/~hy3chan Date: Tuesday, May 08, 2007 Time: 12:00 Location: TTI-C Conference room Title: Notions of Metric Dimension Abstract: The study of finite metrics is an important area of research, because of its wide applications to many different problems. Hence, it is desirable to know what metrics admit better algorithms. Many NP-hard problems become more tractable for low-dimensional Euclidean metrics. However, the notion of Euclidean dimension cannot be applied to general metrics. Clearly, a better notion of dimension should be used to characterize the complexity of a general metric space. One such notion is the doubling dimension, which is popular among the theory community. Many efficient algorithms for problems on low-dimensional Euclidean space have counterparts for metrics with low doubling dimension. Sparse spanners are well studied for Euclidean metrics and we extend the framework to doubling metrics. For instance, we construct linear sized (1 + \eps)-spanners for doubling metrics. Observe that doubling dimension imposes a restriction on the local growth rate everywhere in a metric. Therefore, an otherwise simple metric with a small locally complex region would have high doubling dimension. We investigate a new notion of metric dimension that captures the global behavior of a metric. For metrics with low global dimension, one would expect better guarantees for problems whose objectives depend globally on the input metrics. Indeed, for the Traveling Salesman Problem on such metrics, we give a sub-exponential time algorithm with approximation ratio arbitrarily close to 1. If you have any questions or would like to meet the speaker, please contact Ponda Barnes at pondabarnes@tti-c.org For future TTI-C talks and events please go to http://ttic.uchicago.edu/cal/month.php -------------- next part -------------- An HTML attachment was scrubbed... URL: http://ttic.uchicago.edu/pipermail/colloquium/attachments/20070508/351a0543/attachment-0001.htm From pondabarnes at tti-c.org Tue May 8 15:55:16 2007 From: pondabarnes at tti-c.org (Ponda Barnes) Date: Tue May 8 16:03:40 2007 Subject: FW: [TTIC Colloquium] Guest Speaker Announcement Message-ID: <008101c791b3$2eb81d20$e8bf8780@TTIC47> Skipped content of type multipart/alternative-------------- next part -------------- _______________________________________________ Colloquium mailing list Colloquium@ttic.uchicago.edu http://ttic.uchicago.edu/mailman/listinfo/colloquium From shamkakade at gmail.com Fri May 11 08:33:15 2007 From: shamkakade at gmail.com (Sham Kakade) Date: Fri May 11 08:43:06 2007 Subject: [TTIC Colloquium] Fwd: [statseminars] Statistics Seminar: EMMANUEL CANDES, Monday, May 14, 2007 at 4:00 PM in Eckhart 133 In-Reply-To: <20070508071637.AOM51967@m4500-01.uchicago.edu> References: <20070508071637.AOM51967@m4500-01.uchicago.edu> Message-ID: <2cebf9f0705110633p3f0afec6t96814155f9907bf2@mail.gmail.com> Some people may be interested in the talk below... cheers -Sham ---------- Forwarded message ---------- From: statseminars-admin@listhost.uchicago.edu Date: May 8, 2007 7:16 AM Subject: [statseminars] Statistics Seminar: EMMANUEL CANDES, Monday, May 14, 2007 at 4:00 PM in Eckhart 133 To: statseminars@listhost.uchicago.edu The University of Chicago Department of Statistics Seminar Series EMMANUEL CANDES Department of Applied and Computational Mathematics California Institute of Technology The Dantzig Selector: statistical estimation when p is larger than n MONDAY, May 14, 2007 at 4:00 PM 133 Eckhart Hall, 5734 S. University Avenue Refreshments following the seminar in Eckhart 110. ABSTRACT In many important statistical applications, the number of variables or parameters is much larger than the number of observations. In radiology and biomedical imaging for instance, one is typically able to collect far fewer measurements about an image of interest than the unknown number of pixels. Examples in functional MRI and tomography immediately come to mind. Other examples of high- dimensional data in genomics, signal processing and many other fields abound. In the context of mulitple linear regression for instance, a fundamental question is whether it is possible to estimate a vector of parameters of size p from a vector of observations of size n when n<< p. This seems a priori hopeless. This talk introduces a new estimator, dubbed the "Dantzig selector" in honor of the late George Dantzig as it invokes linear programming, and which enjoys remarkable statistical properties. Suppose that the data or design matrix obeys a uniform uncertainty principle and that the true parameter vector is sufficiently sparse or compressible which roughly guarantees that the model is identifiable. Then the estimator achieves an accuracy which nearly equals that one would achieve with an oracle that would supply perfect information about which coordinates of the unknown parameter vector are nonzero and which were above the noise level. Our results connect with the important model selection problem. In effect, the Dantzig Selector automatically selects the subset of covariates with nearly the best predictive power, by solving a convenient linear program. The results are parts of a larger body of work perhaps best known as "Compressive Sampling" or "Compressed Sensing". If time allows, I will discuss connections with other fields such as signal processing and coding theory. Please send email to Mathias Drton (drton@galton.uchicago.edu) for further information. Information about building access for persons with disabilities may be obtained in advance by calling Karen Gonzalez (Department Administrator and Assistant to Chair) at 773.702.8335 or by email (karen@galton.uchicago.edu). _______________________________________________ statseminars mailing list - statseminars@listhost.uchicago.edu https://listhost.uchicago.edu/mailman/listinfo/statseminars From pondabarnes at tti-c.org Mon May 21 11:37:28 2007 From: pondabarnes at tti-c.org (Ponda Barnes) Date: Mon May 21 11:53:48 2007 Subject: [TTIC Colloquium] Guest Speaker Message-ID: <001901c79bc6$5248c880$e8bf8780@TTIC47> Guest Speaker Announcement Presented by: Toyota Technological Institute at Chicago Speaker: Michael Shub Speaker's home page: http://www.math.utoronto.ca/shub Date: Tuesday, May 22, 2007 Time: 3:00pm Location: TTI-C Conference room Title: Smale's 17th Problem: Recent progress Abstract: In a series of papers written in the first half of the 1990's Steve Smale and I studied the complexity of solving systems of n polynomial equations in n complex variables. We studied path following techniques. A system with known solution is connected by a path to the system we want to solve and the solution is "continued" along the path. The path we chose was the straight line connecting the systems. We proved that "on average" systems can be solved with polynomial cost but we did not prove the existence of a uniform algorithm. The question of the existence of a uniform algorithm is Smale's 17th problem. Recently, Beltran and Pardo have made significant progress on this problem. Moreover, Jointly with Beltran I have linked the complexity to the length of the (problem, solution) path in the condition number Riemannian structure. Surprisingly short paths exist! So the study of the geodesics of this Riemannian structure presents interesting challenges. If you have any questions or would like to meet the speaker, please contact Ponda Barnes at pondabarnes@tti-c.org For future TTI-c talks and events, please visit http://ttic.uchicago.edu/cal/month.php -------------- next part -------------- An HTML attachment was scrubbed... URL: http://ttic.uchicago.edu/pipermail/colloquium/attachments/20070521/d079a2ba/attachment-0001.htm From pondabarnes at tti-c.org Mon May 21 15:18:56 2007 From: pondabarnes at tti-c.org (Ponda Barnes) Date: Mon May 21 15:35:23 2007 Subject: [TTIC Colloquium] Guest Speaker Announcement Message-ID: <005101c79be5$44d00cd0$e8bf8780@TTIC47> Guest Speaker Presented by: Toyota Technological Institute at Chicago Speaker: Greg Mori Speaker's home page: http://www.cs.sfu.ca/~mori/ Date: Thursday, May 24, 2007 Time: 10:00 am Location: TTI-C Conference room Title: Detecting Pedestrians by Learning Shapelet Features and Boosted Multiple Deformable Trees for Parsing Human Poses Abstract: In this talk, we present two pieces of work in the "Looking at People" domain. In the first part, we address the problem of detecting pedestrians in still images. We introduce an algorithm for learning shapelet features, a set of mid-level features. These features are focused on local regions of the image and are built from low-level gradient information that discriminates between pedestrian and non-pedestrian classes. Using AdaBoost, these shapelet features are created as a combination of oriented gradient responses. To train the final classifier, we use AdaBoost for a second time to select a subset of our learned shapelets. By first focusing locally on smaller feature sets, our algorithm attempts to harvest more useful information than by examining all the low-level features together. We present quantitative results demonstrating the effectiveness of our algorithm. In particular, we obtain an error rate 14 percentage points lower (at $10^ {-6} $ FPPW) than the previous state of the art detector of Dalal and Triggs on the INRIA dataset. In the second part, we present a method for estimating human pose in still images. Tree-structured models have been widely used for this problem. While such models allow efficient learning and inference, they fail to capture additional dependencies between body parts, other than kinematic constraints. In this paper, we consider the use of multiple tree models, rather than a singletree model for human pose estimation. Our model can alleviate the limitations of a single tree-structured model by combining information provided across different tree models. The parameters of each individual tree model are trained via standard learning algorithms in a single tree-structured model. Different tree models are combined in a discriminative fashion by a boosting procedure. We present experimental results showing the improvement of our model over previous approaches on a very challenging dataset. If you have any questions or would like to meet the speaker, please contact Ponda Barnes at pondabarnes@tti-c.org For future TTI-C talks and events, please visit http://ttic.uchicago.edu/cal/month.php -------------- next part -------------- An HTML attachment was scrubbed... URL: http://ttic.uchicago.edu/pipermail/colloquium/attachments/20070521/4d4caf08/attachment.htm From pondabarnes at tti-c.org Tue May 22 14:56:19 2007 From: pondabarnes at tti-c.org (Ponda Barnes) Date: Tue May 22 15:13:25 2007 Subject: [TTIC Colloquium] FW: Guest Speaker Message-ID: <003701c79cab$499cec30$e8bf8780@TTIC47> Reminder!! Guest Speaker Announcement Presented by: Toyota Technological Institute at Chicago Speaker: Michael Shub Speaker's home page: http://www.math.utoronto.ca/shub Date: Tuesday, May 22, 2007 Time: 3:00pm Location: TTI-C Conference room Title: Smale's 17th Problem: Recent progress Abstract: In a series of papers written in the first half of the 1990's Steve Smale and I studied the complexity of solving systems of n polynomial equations in n complex variables. We studied path following techniques. A system with known solution is connected by a path to the system we want to solve and the solution is "continued" along the path. The path we chose was the straight line connecting the systems. We proved that "on average" systems can be solved with polynomial cost but we did not prove the existence of a uniform algorithm. The question of the existence of a uniform algorithm is Smale's 17th problem. Recently, Beltran and Pardo have made significant progress on this problem. Moreover, Jointly with Beltran I have linked the complexity to the length of the (problem, solution) path in the condition number Riemannian structure. Surprisingly short paths exist! So the study of the geodesics of this Riemannian structure presents interesting challenges. If you have any questions or would like to meet the speaker, please contact Ponda Barnes at pondabarnes@tti-c.org For future TTI-c talks and events, please visit http://ttic.uchicago.edu/cal/month.php -------------- next part -------------- An HTML attachment was scrubbed... URL: http://ttic.uchicago.edu/pipermail/colloquium/attachments/20070522/4dafb9c4/attachment-0001.htm From pondabarnes at tti-c.org Tue May 22 14:57:26 2007 From: pondabarnes at tti-c.org (Ponda Barnes) Date: Tue May 22 15:14:21 2007 Subject: [TTIC Colloquium] FW: Guest Speaker Announcement Message-ID: <003c01c79cab$6be52af0$e8bf8780@TTIC47> REMINDER!! Guest Speaker Presented by: Toyota Technological Institute at Chicago Speaker: Greg Mori Speaker's home page: http://www.cs.sfu.ca/~mori/ Date: Thursday, May 24, 2007 Time: 10:00 am Location: TTI-C Conference room Title: Detecting Pedestrians by Learning Shapelet Features and Boosted Multiple Deformable Trees for Parsing Human Poses Abstract: In this talk, we present two pieces of work in the "Looking at People" domain. In the first part, we address the problem of detecting pedestrians in still images. We introduce an algorithm for learning shapelet features, a set of mid-level features. These features are focused on local regions of the image and are built from low-level gradient information that discriminates between pedestrian and non-pedestrian classes. Using AdaBoost, these shapelet features are created as a combination of oriented gradient responses. To train the final classifier, we use AdaBoost for a second time to select a subset of our learned shapelets. By first focusing locally on smaller feature sets, our algorithm attempts to harvest more useful information than by examining all the low-level features together. We present quantitative results demonstrating the effectiveness of our algorithm. In particular, we obtain an error rate 14 percentage points lower (at $10^ {-6} $ FPPW) than the previous state of the art detector of Dalal and Triggs on the INRIA dataset. In the second part, we present a method for estimating human pose in still images. Tree-structured models have been widely used for this problem. While such models allow efficient learning and inference, they fail to capture additional dependencies between body parts, other than kinematic constraints. In this paper, we consider the use of multiple tree models, rather than a singletree model for human pose estimation. Our model can alleviate the limitations of a single tree-structured model by combining information provided across different tree models. The parameters of each individual tree model are trained via standard learning algorithms in a single tree-structured model. Different tree models are combined in a discriminative fashion by a boosting procedure. We present experimental results showing the improvement of our model over previous approaches on a very challenging dataset. If you have any questions or would like to meet the speaker, please contact Ponda Barnes at pondabarnes@tti-c.org For future TTI-C talks and events, please visit http://ttic.uchicago.edu/cal/month.php -------------- next part -------------- An HTML attachment was scrubbed... URL: http://ttic.uchicago.edu/pipermail/colloquium/attachments/20070522/3620d866/attachment.htm