From pondabarnes at tti-c.org Thu Mar 1 08:36:35 2007 From: pondabarnes at tti-c.org (Ponda Barnes) Date: Thu Mar 1 08:46:23 2007 Subject: [TTIC Colloquium] Guest Speaker Message-ID: <200703011446.l21EkK2m027612@nagoya.uchicago.edu> TTI-C Guest Speaker Presented by: Toyota Technological Institute at Chicago Speaker: Satyen Kale Speaker's home page: http://www.cs.princeton.edu/~satyen Date: Thursday, March 01, 2007 Location: TTI-Conference room Time: 10:00 Title: Efficient Algorithms using the Multiplicative Weights Update Method Abstract: Algorithms based on convex optimization, especially linear and semidefinite programming, are ubiquitous in Computer Science. While there are polynomial time algorithms known to solve such problems, quite often the running time of these algorithms is very high. Designing simpler and more efficient algorithms is important for practical impact. In my talk, I will describe applications of a Lagrangian relaxation technique, the Multiplicative Weights Update method, in the design of efficient algorithms for various optimization problems. We generalize the method to the setting of symmetric matrices rather than real numbers. The new algorithm yields the first truly general, combinatorial, primal-dual method for designing efficient algorithms using semidefinite programming. Using these techniques, we obtain significantly faster algorithms for approximating the Sparsest Cut and Balanced Separator in both directed and undirected weighted graphs, and the Min UnCut problem. In addition, we also obtain an efficient derandomization of the Alon-Roichman theorem, a deterministic O(log n) approximation to the Quantum Hypergraph Covering problem, and an alternative proof of Aaronson's result on the learnability of quantum states. 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/20070301/372f8120/attachment-0001.htm From pondabarnes at tti-c.org Thu Mar 1 08:54:02 2007 From: pondabarnes at tti-c.org (Ponda Barnes) Date: Thu Mar 1 09:03:54 2007 Subject: [TTIC Colloquium] Distinguished Lecture Message-ID: <200703011503.l21F3r2m027689@nagoya.uchicago.edu> TTI-C Distinguished Lecture Presented by: Toyota Technological Institute at Chicago Speaker: John Lafferty Speaker's home page: http://www.cs.cmu.edu/~lafferty Date: Wednesday, March 7, 2007 Location: Kent, room 120 Time: 2:00 pm Title: Challenges in Statistical Machine Learning Abstract: A surge of research in machine learning during the past decade has led to powerful learning methods that are successfully being applied to a wide range of application domains, from search engines to computational biology and robotics. These advances have in part been achieved by refining the art and engineering practice of machine learning, paralleled by a confluence of machine learning and statistics. But an understanding of the scientific foundations and fundamental limits to learning from data can also be effectively leveraged in practice. In this overview of recent work we present some of the current technical challenges in the field of machine learning, focusing on high dimensional data and minimax rates of convergence, a measure of learnability that parallels channel capacity in information theory. These challenges include understanding the role of sparsity in statistical learning, semi-supervised learning, the tradeoff between computation and risk, and structured prediction 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/20070301/4cae057e/attachment.htm From pondabarnes at tti-c.org Mon Mar 5 09:36:58 2007 From: pondabarnes at tti-c.org (Ponda Barnes) Date: Mon Mar 5 09:49:25 2007 Subject: [TTIC Colloquium] Guest Speaker Message-ID: <200703051549.l25FnM2m016687@nagoya.uchicago.edu> TTI-C Guest Speaker Presented by: Toyota Technological Institute at Chicago Speaker: Kilian Weinberger Speaker's home page: http://www.seas.upenn.edu/~kilianw/kqw/Welcome.html Date: Tuesday, March 6, 2007 Location: TTI-C Conference room Time: 10:00 Title: Metric Learning with Convex Optimization Abstract: Many problems in computer science can be simplified by clever representations of sensory or symbolic input. How to discover such representations automatically, from large amounts of data, remains a fundamental challenge. The goal of metric learning is to derive Euclidean representations of labeled or unlabeled inputs from observed statistical regularities. In this talk I will review two recently proposed algorithms for metric learning. Both algorithms rely on modern tools in convex optimization that are proving increasingly useful in many areas of machine learning. 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/20070305/88a9d9f4/attachment.htm From pondabarnes at tti-c.org Tue Mar 6 08:02:06 2007 From: pondabarnes at tti-c.org (Ponda Barnes) Date: Tue Mar 6 08:15:08 2007 Subject: [TTIC Colloquium] FW: Guest Speaker Message-ID: <200703061415.l26EF52m020208@nagoya.uchicago.edu> REMINDER! TTI-C Guest Speaker Presented by: Toyota Technological Institute at Chicago Speaker: Kilian Weinberger Speaker's home page: http://www.seas.upenn.edu/~kilianw/kqw/Welcome.html Date: Tuesday, March 6, 2007 Location: TTI-C Conference room Time: 10:00 Title: Metric Learning with Convex Optimization Abstract: Many problems in computer science can be simplified by clever representations of sensory or symbolic input. How to discover such representations automatically, from large amounts of data, remains a fundamental challenge. The goal of metric learning is to derive Euclidean representations of labeled or unlabeled inputs from observed statistical regularities. In this talk I will review two recently proposed algorithms for metric learning. Both algorithms rely on modern tools in convex optimization that are proving increasingly useful in many areas of machine learning. 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/20070306/7c6210cf/attachment-0001.htm From pondabarnes at tti-c.org Tue Mar 6 08:04:11 2007 From: pondabarnes at tti-c.org (Ponda Barnes) Date: Tue Mar 6 08:17:08 2007 Subject: [TTIC Colloquium] FW: Distinguished Lecture Message-ID: <200703061417.l26EH62m020274@nagoya.uchicago.edu> REMINDER! TTI-C Distinguished Lecture Presented by: Toyota Technological Institute at Chicago Speaker: John Lafferty Speaker's home page: http://www.cs.cmu.edu/~lafferty Date: Wednesday, March 7, 2007 Location: Kent, room 120 Time: 2:00 pm Title: Challenges in Statistical Machine Learning Abstract: A surge of research in machine learning during the past decade has led to powerful learning methods that are successfully being applied to a wide range of application domains, from search engines to computational biology and robotics. These advances have in part been achieved by refining the art and engineering practice of machine learning, paralleled by a confluence of machine learning and statistics. But an understanding of the scientific foundations and fundamental limits to learning from data can also be effectively leveraged in practice. In this overview of recent work we present some of the current technical challenges in the field of machine learning, focusing on high dimensional data and minimax rates of convergence, a measure of learnability that parallels channel capacity in information theory. These challenges include understanding the role of sparsity in statistical learning, semi-supervised learning, the tradeoff between computation and risk, and structured prediction 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/20070306/1d8f1dbf/attachment.htm From pondabarnes at tti-c.org Wed Mar 7 13:27:14 2007 From: pondabarnes at tti-c.org (Ponda Barnes) Date: Wed Mar 7 13:40:58 2007 Subject: [TTIC Colloquium] Guest Speaker Message-ID: <200703071940.l27Jeu2m024915@nagoya.uchicago.edu> TTI-C Guest Speaker Presented by: Toyota Technological Institute at Chicago Speaker: Shai Shalev-Schwartz Speaker's home page: http://seminars.ijs.si/pascal/2005/nips05_whistler/video.asp?video_id=1196 Date: Thursday, March 8, 2007 Location: TTI-C Conference room Time: 12:30 Title: Convex Repeated Games, Regret, and Duality. Abstract: We describe an algorithmic framework for an abstract game, which we term a convex repeated game. We show that various machine-learning algorithms for online learning and boosting can be all derived as special cases of our algorithmic framework. This unified view explains the properties of existing algorithms and also enables us to derive several new interesting algorithms. Our algorithmic framework stems from a connection that we build between the notions of regret in game theory and weak duality in convex optimization. We also underscore the applicability of convex repeated games and the derived algorithmic framework for stochastic optimization, and game theory. 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/20070307/996f399a/attachment.htm From pondabarnes at tti-c.org Thu Mar 8 09:38:16 2007 From: pondabarnes at tti-c.org (Ponda Barnes) Date: Thu Mar 8 09:52:31 2007 Subject: [TTIC Colloquium] FW: Guest Speaker Message-ID: <200703081552.l28FqT2m027508@nagoya.uchicago.edu> REMINDER!! TTI-C Guest Speaker Presented by: Toyota Technological Institute at Chicago Speaker: Shai Shalev-Schwartz Speaker's home page: http://seminars.ijs.si/pascal/2005/nips05_whistler/video.asp?video_id=1196 Date: Thursday, March 8, 2007 Location: TTI-C Conference room Time: 12:30 Title: Convex Repeated Games, Regret, and Duality. Abstract: We describe an algorithmic framework for an abstract game, which we term a convex repeated game. We show that various machine-learning algorithms for online learning and boosting can be all derived as special cases of our algorithmic framework. This unified view explains the properties of existing algorithms and also enables us to derive several new interesting algorithms. Our algorithmic framework stems from a connection that we build between the notions of regret in game theory and weak duality in convex optimization. We also underscore the applicability of convex repeated games and the derived algorithmic framework for stochastic optimization, and game theory. 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/20070308/e36430e3/attachment-0001.htm From pondabarnes at tti-c.org Mon Mar 12 08:27:26 2007 From: pondabarnes at tti-c.org (Ponda Barnes) Date: Mon Mar 12 09:28:37 2007 Subject: [TTIC Colloquium] Guest Speaker Message-ID: <200703121528.l2CFSY2m015857@nagoya.uchicago.edu> TTI-C Guest Speaker Presented by: Toyota Technological Institute at Chicago Speaker: Jia Limin Speaker's home page: http://www.cs.princeton.edu/~ljia/ Date: Tuesday, March 13, 2007 Location: TTI-C Conference room Time: 10:00am Title: Linear Logic and Imperative Programming Abstract: Improper pointer operations in software written in programming languages such as C significantly compromise the reliability and security of software systems. Therefore, one of the most important and enduring problems in programming languages research involves verification of programs that construct, manipulate, and dispose of complex heap-allocated data structures. In my talk, I will first present our new sub structural logic for reasoning about the memory safety of imperative programs. Our new logic modularly integrates sub structural reasoning with constraint-based reasoning such as integer constraints. Next, I will introduce a new imperative language that allows programmers to define and manipulate recursive data structures using formulas in our logic. By combining new verification techniques based on sub structural logics with a modern type system for resource control, this language guarantees not only the memory safety of operations such as deallocation and dereferencing, but also ensures that shape invariants of data structures hold throughout the execution of the program. 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/20070312/edb50148/attachment.htm From pondabarnes at tti-c.org Mon Mar 12 08:36:47 2007 From: pondabarnes at tti-c.org (Ponda Barnes) Date: Mon Mar 12 09:37:57 2007 Subject: [TTIC Colloquium] Guest speaker Message-ID: <200703121537.l2CFbu2m015897@nagoya.uchicago.edu> TTI-C Guest Speaker Presented by: Toyota Technological Institute at Chicago Speaker: Yury Makarychev Speaker's home page: http://www.cs.princeton.edu/~ymakaryc/ Date: Thursday, March 15, 2007 Location: TTI-C Conference room Time: 10:00am Title: Approximation Algorithms for Unique Games Abstract: Unique games are constraint satisfaction problems that can be viewed as a generalization of MAX CUT to a larger domain: We are given a graph G = (V,E) on n vertices and a permutation P_{uv} on the set of labels {1,...,k} for every edge (u, v). Our goal is to assign a label X_u in {1,..., k} to each vertex u, so as to maximize the number of satisfied constraints P_{uv} (X_u) = X_v. This problem has recently attracted a lot of attention since hardness of approximation for many problems, such as Sparsest Cut and Vertex Cover, was proved assuming the Unique Games Conjecture. Roughly speaking, this conjecture says that even if almost all constraints in a unique game are satisfiable it is NP-hard to satisfy a small constant fraction of constraints. Unique games pose a great challenge for our existing techniques: Typically, semidefinite programming (SDP) relaxations are well suited for optimization problems involving boolean variables (e.g. MAX CUT). But little is known about how to analyze SDP solutions for problems with larger domains. We present three approximation algorithms for Unique Games that satisfy roughly k^ {-epsilon/2}, 1 - O (sqrt{epsilon log k}) and 1 - epsilon * O(sqrt{log k log n}) fraction of all constraints if a (1-epsilon) fraction of all constraints is satisfiable. This talk is based on joint papers with Moses Charikar, Eden Chlamtac, and Konstantin Makaryche . 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/20070312/e047cbf0/attachment.htm From pondabarnes at tti-c.org Tue Mar 13 08:03:01 2007 From: pondabarnes at tti-c.org (Ponda Barnes) Date: Tue Mar 13 09:04:50 2007 Subject: [TTIC Colloquium] FW: Guest Speaker Message-ID: <200703131504.l2DF4l2m020061@nagoya.uchicago.edu> Reminder!! TTI-C Guest Speaker Presented by: Toyota Technological Institute at Chicago Speaker: Jia Limin Speaker's home page: http://www.cs.princeton.edu/~ljia/ Date: Tuesday, March 13, 2007 Location: TTI-C Conference room Time: 10:00am Title: Linear Logic and Imperative Programming Abstract: Improper pointer operations in software written in programming languages such as C significantly compromise the reliability and security of software systems. Therefore, one of the most important and enduring problems in programming languages research involves verification of programs that construct, manipulate, and dispose of complex heap-allocated data structures. In my talk, I will first present our new sub structural logic for reasoning about the memory safety of imperative programs. Our new logic modularly integrates sub structural reasoning with constraint-based reasoning such as integer constraints. Next, I will introduce a new imperative language that allows programmers to define and manipulate recursive data structures using formulas in our logic. By combining new verification techniques based on sub structural logics with a modern type system for resource control, this language guarantees not only the memory safety of operations such as deallocation and dereferencing, but also ensures that shape invariants of data structures hold throughout the execution of the program. 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/20070313/190bb418/attachment-0001.htm From pondabarnes at tti-c.org Thu Mar 15 07:59:56 2007 From: pondabarnes at tti-c.org (Ponda Barnes) Date: Thu Mar 15 09:02:34 2007 Subject: [TTIC Colloquium] FW: Guest speaker Message-ID: <200703151502.l2FF2V2m028805@nagoya.uchicago.edu> Reminder!! TTI-C Guest Speaker Presented by: Toyota Technological Institute at Chicago Speaker: Yury Makarychev Speaker's home page: http://www.cs.princeton.edu/~ymakaryc/ Date: Thursday, March 15, 2007 Location: TTI-C Conference room Time: 10:00am Title: Approximation Algorithms for Unique Games Abstract: Unique games are constraint satisfaction problems that can be viewed as a generalization of MAX CUT to a larger domain: We are given a graph G = (V,E) on n vertices and a permutation P_{uv} on the set of labels {1,...,k} for every edge (u, v). Our goal is to assign a label X_u in {1,..., k} to each vertex u, so as to maximize the number of satisfied constraints P_{uv} (X_u) = X_v. This problem has recently attracted a lot of attention since hardness of approximation for many problems, such as Sparsest Cut and Vertex Cover, was proved assuming the Unique Games Conjecture. Roughly speaking, this conjecture says that even if almost all constraints in a unique game are satisfiable it is NP-hard to satisfy a small constant fraction of constraints. Unique games pose a great challenge for our existing techniques: Typically, semidefinite programming (SDP) relaxations are well suited for optimization problems involving boolean variables (e.g. MAX CUT). But little is known about how to analyze SDP solutions for problems with larger domains. We present three approximation algorithms for Unique Games that satisfy roughly k^ {-epsilon/2}, 1 - O (sqrt{epsilon log k}) and 1 - epsilon * O(sqrt{log k log n}) fraction of all constraints if a (1-epsilon) fraction of all constraints is satisfiable. This talk is based on joint papers with Moses Charikar, Eden Chlamtac, and Konstantin Makaryche . 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/20070315/bc389d3c/attachment.htm From pondabarnes at tti-c.org Fri Mar 16 16:01:44 2007 From: pondabarnes at tti-c.org (Ponda Barnes) Date: Fri Mar 16 17:05:24 2007 Subject: [TTIC Colloquium] Guest Speaker Message-ID: <200703162305.l2GN5L2m000679@nagoya.uchicago.edu> Guest Speaker announcement Presented by: Toyota Technological Institute at Chicago Speaker: Yuan QI Speaker's home page: http://web.media.mit.edu/~yuanqi/ Date: Tuesday, March 20, 2007 Time: 10:00am Location: TTI-C Conference room Title: Bayesian Learning for Deciphering Gene Regulation Abstract: Gene regulation plays a fundamental role in biological systems. As more high-throughput biological data becomes available it is possible to quantitatively study gene regulation in a systematic way. In this talk I present my work on three related problems on gene regulation including: (1) identifying genes that affect organism development; (2) detecting protein-DNA binding events and cis-regulatory elements; (3) and deciphering regulatory cascades at the transcriptional level for embryonic stem cell development. To address these problems, we must overcome many computational challenges, including little prior biological knowledge, joint effect of many biological variables, and large model spaces for learning. Facing these computational challenges, I developed novel Bayesian methods to analyze high-throughput data, in order to deepen our understanding of gene regulation for organism development. Specifically, I first devised a novel Bayesian semi-supervised classification method to identify candidate genes specific to certain lineages and cell-types of C. elegans embryos. My computational predictions about some previously uncharacterized genes were experimentally confirmed by my biologist collaborators. Second, I built a new Bayesian graphical model of protein-DNA binding and developed an approximate inference algorithm to efficiently estimate binding events in high spatial-resolution and guide motif discovery. The software implementation of this algorithm is being used by research groups worldwide. Third, I developed a novel nonparametric Bayesian model that enables the reconstruction of a regulatory cascade for the development of embryonic stem cells, without predefining the level of the cascade or the branching number at each level. Some predictions were experimentally confirmed by our collaborators and independently by other research groups. 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/20070316/24c971d2/attachment-0001.htm From pondabarnes at tti-c.org Tue Mar 20 07:35:04 2007 From: pondabarnes at tti-c.org (Ponda Barnes) Date: Tue Mar 20 08:40:55 2007 Subject: [TTIC Colloquium] FW: Guest Speaker Message-ID: <200703201440.l2KEeq2m030407@nagoya.uchicago.edu> Reminder!! Guest Speaker announcement Presented by: Toyota Technological Institute at Chicago Speaker: Yuan QI Speaker's home page: http://web.media.mit.edu/~yuanqi/ Date: Tuesday, March 20, 2007 Time: 10:00am Location: TTI-C Conference room Title: Bayesian Learning for Deciphering Gene Regulation Abstract: Gene regulation plays a fundamental role in biological systems. As more high-throughput biological data becomes available it is possible to quantitatively study gene regulation in a systematic way. In this talk I present my work on three related problems on gene regulation including: (1) identifying genes that affect organism development; (2) detecting protein-DNA binding events and cis-regulatory elements; (3) and deciphering regulatory cascades at the transcriptional level for embryonic stem cell development. To address these problems, we must overcome many computational challenges, including little prior biological knowledge, joint effect of many biological variables, and large model spaces for learning. Facing these computational challenges, I developed novel Bayesian methods to analyze high-throughput data, in order to deepen our understanding of gene regulation for organism development. Specifically, I first devised a novel Bayesian semi-supervised classification method to identify candidate genes specific to certain lineages and cell-types of C. elegans embryos. My computational predictions about some previously uncharacterized genes were experimentally confirmed by my biologist collaborators. Second, I built a new Bayesian graphical model of protein-DNA binding and developed an approximate inference algorithm to efficiently estimate binding events in high spatial-resolution and guide motif discovery. The software implementation of this algorithm is being used by research groups worldwide. Third, I developed a novel nonparametric Bayesian model that enables the reconstruction of a regulatory cascade for the development of embryonic stem cells, without predefining the level of the cascade or the branching number at each level. Some predictions were experimentally confirmed by our collaborators and independently by other research groups. 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/20070320/8a5d9027/attachment.htm From pondabarnes at tti-c.org Tue Mar 20 11:57:58 2007 From: pondabarnes at tti-c.org (Ponda Barnes) Date: Tue Mar 20 13:04:00 2007 Subject: [TTIC Colloquium] Guest Speaker announcement Message-ID: <200703201903.l2KJ3t2m031422@nagoya.uchicago.edu> Guest Speaker Presented by: Toyota Technological Institute at Chicago Speaker: Sergey Yekhanin Speaker's home page: http://theory.lcs.mit.edu/~yekhanin/ Date: Wednesday, March 28, 2007 Time: 10:00am Location: TTI-C Conference room Title: New Locally Decodable Codes and Private Information Retrieval Schemes Abstract: A q-query Locally Decodable Code (LDC) is an error-correcting code that encodes an n-bit message x as a codeword C(x), such that one can probabilistically recover any bit x_i of the message by querying only q bits of the codeword C(x), even after some constant fraction of codeword bits has been corrupted. The goal of LDC related research is to minimize the length of such codes. A q-server private information retrieval (PIR) scheme is a an n-bit string x replicated between q servers while each server individually learns no information about i. The goal of PIR related research is to minimize the communication complexity of such schemes. We present a novel algebraic approach to LDCs and PIRs and obtain vast improvements upon the earlier work. Specifically, given any Mersenne prime p = 2^t ? 1, we design three query LDCs of length Exp (n^{1/t}), for every n. Based on the largest known Mersenne prime, this translates to a length of less than Exp(n^{10^{?7}}), compared to Exp(n^{1/2}) in the previous constructions. We also design 3-server PIR schemes with communication complexity of O(n^{10^{?7}}) to access an n-bit database, compared to the previous best scheme with complexity O(n^{1/5.25}). It has often been conjectured that there are infinitely many Mersenne primes. Under this conjecture, our constructions yield three query locally decodable codes of sub exponential length and three server private information retrieval schemes with sub polynomial communication complexity. 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/20070320/aa9fc580/attachment-0001.htm From pondabarnes at tti-c.org Wed Mar 21 07:21:23 2007 From: pondabarnes at tti-c.org (Ponda Barnes) Date: Wed Mar 21 08:27:49 2007 Subject: [TTIC Colloquium] Guest Speaker announcement Message-ID: <200703211427.l2LERl2m006089@nagoya.uchicago.edu> Guest Speaker Presented by: Toyota Technological Institute at Chicago Speaker: David Woodruff Speaker's home page: http://web.mit.edu/dpwood/www/ Date: 3/21/07 Time: 10:00am Location: TTI-C Conference room Title: Efficient and Private Distance Approximation Abstract: I will cover two of my results in distance approximation. Consider the setting in which two parties want to approximate the distance between their input vectors. First, I will consider l_2, the Euclidean distance. It is known how to approximate l_2 efficiently. However, if we require the protocol to be private, that is, neither party can learn more than what follows from the distance and his/her private input, much less is known. Feigenbaum, Ishai, Malkin, Nissim, Strauss, and Wright [FIMNSW] gave a protocol with O(sqrt{d}) communication for privately approximating the Hamming distance of two d-dimensional vectors. I will give a private protocol with polylog(d) communication for l_2. As a special case, this yields an exponential improvement over [FIMNSW] for the Hamming distance. Next I will consider the l_p distance, for p > 2. This problem is motivated by recent research in streaming algorithms. I will give a 1-round protocol achieving optimal communication for this problem, up to logarithmic factors. It can be implemented in the streaming model, and consequently resolves one of the main open questions of a 1996 paper of Alon, Matias, and Szegedy. Joint work with Piotr Indyk. 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/20070321/10907d38/attachment.htm From pondabarnes at tti-c.org Mon Mar 26 08:17:47 2007 From: pondabarnes at tti-c.org (Ponda Barnes) Date: Mon Mar 26 09:27:19 2007 Subject: [TTIC Colloquium] Guest Speaker Announcement Message-ID: <200703261527.l2QFRG2m020162@nagoya.uchicago.edu> Guest Speaker Presented by: Toyota Technological Institute at Chicago Speaker: Greg Shakhnarovich Speaker's home page: http://people.csail.mit.edu/gregory/ Date: Tuesday, March 27, 2007 Time: 10:00 am Location: TTI-C Conference room Title: Learning Task-Specific Similarity Abstract: The notion of similarity is fundamental in machine learning. The ability to assess similarity, and to find in a database examples similar to a given instance, is central to many statistical learning methods. Similarity is commonly modeled in terms of a distance function in the input space. However, such a definition may not capture the concept of similarity relevant to the task at hand. In this talk, I will describe an approach to learning similarity from user-provided examples of what is deemed similar (and, optionally, dissimilar). I will show how to construct, using a greedy algorithm inspired by boosting, an embedding of the data into a weighted binary space, where a simple metric approximates the original similarity concept. This approach provides a means for data compression focused on preserving relevant features, and enables efficient search with respect to user-defined similarity in very large databases. I will show how this makes example-based learning work well in some of the problems in computer vision for which it was previously deemed infeasible. 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/20070326/abf836ed/attachment-0001.htm From pondabarnes at tti-c.org Tue Mar 27 08:52:29 2007 From: pondabarnes at tti-c.org (Ponda Barnes) Date: Tue Mar 27 10:02:39 2007 Subject: [TTIC Colloquium] FW: Guest Speaker Announcement Message-ID: <200703271602.l2RG2b2m028313@nagoya.uchicago.edu> Reminder!! Guest Speaker Presented by: Toyota Technological Institute at Chicago Speaker: Greg Shakhnarovich Speaker's home page: http://people.csail.mit.edu/gregory/ Date: Tuesday, March 27, 2007 Time: 10:00 am Location: TTI-C Conference room Title: Learning Task-Specific Similarity Abstract: The notion of similarity is fundamental in machine learning. The ability to assess similarity, and to find in a database examples similar to a given instance, is central to many statistical learning methods. Similarity is commonly modeled in terms of a distance function in the input space. However, such a definition may not capture the concept of similarity relevant to the task at hand. In this talk, I will describe an approach to learning similarity from user-provided examples of what is deemed similar (and, optionally, dissimilar). I will show how to construct, using a greedy algorithm inspired by boosting, an embedding of the data into a weighted binary space, where a simple metric approximates the original similarity concept. This approach provides a means for data compression focused on preserving relevant features, and enables efficient search with respect to user-defined similarity in very large databases. I will show how this makes example-based learning work well in some of the problems in computer vision for which it was previously deemed infeasible. 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/20070327/96d8891d/attachment.htm From pondabarnes at tti-c.org Tue Mar 27 15:43:09 2007 From: pondabarnes at tti-c.org (Ponda Barnes) Date: Tue Mar 27 16:53:31 2007 Subject: [TTIC Colloquium] Guest speaker announcement Message-ID: <200703272253.l2RMrS2m029253@nagoya.uchicago.edu> Guest Speaker Presented by: Toyota Technological Institute at Chicago Speaker: Sergey Yekhanin Speaker's home page: http://theory.ics.mit.edu/~yekhanin/ Date: Wednesday, March 28, 2007 Time: 10:00 Location: TTI-C Conference room Title: New Locally Decodable Codes and Private Information Retrieval Schemes Abstract: A q-query Locally Decodable Code (LDC) is an error-correcting code that encodes an n-bit message x as a codeword C(x), such that one can probabilistically recover any bit x_i of the message by querying only q bits of the codeword C(x), even after some constant fraction of codeword bits has been corrupted. The goal of LDC related research is to minimize the length of such codes. A q-server private information retrieval (PIR) scheme is a an n-bit string x replicated between q servers while each server individually learns no information about i. The goal of PIR related research is to minimize the communication complexity of such schemes. We present a novel algebraic approach to LDCs and PIRs and obtain vast improvements upon the earlier work. Specifically, given any Mersenne prime p = 2^t? 1, we design three query LDCs of length Exp(n^{1/t}), for every n. Based on the largest known Mersenne prime, this translates to a length of less than Exp(n^{10^{?7}}), compared to Exp(n^{1/2}) in the previous constructions. We also design 3-server PIR schemes with communication complexity of O(n^{10^{?7}}) to access an n-bit database, compared to the previous best scheme with complexity O(n^{1/5.25}). It has often been conjectured that there are infinitely many Mersenne primes. Under this conjecture, our constructions yield three query locally decodable codes of sub exponential length and three servers private information retrieval schemes with subpolynomial communication complexity. 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/20070327/f0a90e9d/attachment-0001.htm From pondabarnes at tti-c.org Wed Mar 28 08:42:12 2007 From: pondabarnes at tti-c.org (Ponda Barnes) Date: Wed Mar 28 09:53:06 2007 Subject: [TTIC Colloquium] FW: Guest speaker announcement Message-ID: <200703281553.l2SFr32m031502@nagoya.uchicago.edu> REMINDER!! Guest Speaker Presented by: Toyota Technological Institute at Chicago Speaker: Sergey Yekhanin Speaker's home page: http://theory.ics.mit.edu/~yekhanin/ Date: Wednesday, March 28, 2007 Time: 10:00 Location: TTI-C Conference room Title: New Locally Decodable Codes and Private Information Retrieval Schemes Abstract: A q-query Locally Decodable Code (LDC) is an error-correcting code that encodes an n-bit message x as a codeword C(x), such that one can probabilistically recover any bit x_i of the message by querying only q bits of the codeword C(x), even after some constant fraction of codeword bits has been corrupted. The goal of LDC related research is to minimize the length of such codes. A q-server private information retrieval (PIR) scheme is a an n-bit string x replicated between q servers while each server individually learns no information about i. The goal of PIR related research is to minimize the communication complexity of such schemes. We present a novel algebraic approach to LDCs and PIRs and obtain vast improvements upon the earlier work. Specifically, given any Mersenne prime p = 2^t? 1, we design three query LDCs of length Exp(n^{1/t}), for every n. Based on the largest known Mersenne prime, this translates to a length of less than Exp(n^{10^{?7}}), compared to Exp(n^{1/2}) in the previous constructions. We also design 3-server PIR schemes with communication complexity of O(n^{10^{?7}}) to access an n-bit database, compared to the previous best scheme with complexity O(n^{1/5.25}). It has often been conjectured that there are infinitely many Mersenne primes. Under this conjecture, our constructions yield three query locally decodable codes of sub exponential length and three servers private information retrieval schemes with subpolynomial communication complexity. 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/20070328/ff081c9a/attachment.htm From pondabarnes at tti-c.org Wed Mar 28 13:25:13 2007 From: pondabarnes at tti-c.org (Ponda Barnes) Date: Wed Mar 28 14:35:54 2007 Subject: [TTIC Colloquium] Guest speaker announcement Message-ID: <200703282035.l2SKZq2m031964@nagoya.uchicago.edu> Guest Speaker Presented by: Toyota Technological Institute at Chicago Speaker: Konstantin Makarychev Speaker's home page: http://www.cs.princeton.edu/~kmakaryc/ Date: Thursday, March 29, 2007 Time: 10:00 Location: TTI-C Conference room Title: Quadratic Forms on Graphs Abstract: ============================================ We introduce a new graph parameter, called the Grothendieck constant of a graph, which measures the integrality gap of certain integer programs, and has applications to various combinatorial optimization problems. Its study leads to several extensions of the classical inequality of Grothendieck, to an improvement of a recent result of Kashin and Szarek, and to a solution of a problem of Megretski and of Charikar and Wirth. Most of the talk is based on a joint work with Noga Alon, Yury Makarychev and Assaf Naor. 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/20070328/025ce650/attachment.htm From pondabarnes at tti-c.org Thu Mar 29 08:50:10 2007 From: pondabarnes at tti-c.org (Ponda Barnes) Date: Thu Mar 29 10:01:26 2007 Subject: [TTIC Colloquium] FW: Guest speaker announcement Message-ID: <200703291601.l2TG1N2m001716@nagoya.uchicago.edu> REMINDER!! Guest Speaker Presented by: Toyota Technological Institute at Chicago Speaker: Konstantin Makarychev Speaker's home page: http://www.cs.princeton.edu/~kmakaryc/ Date: Thursday, March 29, 2007 Time: 10:00 Location: TTI-C Conference room Title: Quadratic Forms on Graphs Abstract: ============================================ We introduce a new graph parameter, called the Grothendieck constant of a graph, which measures the integrality gap of certain integer programs, and has applications to various combinatorial optimization problems. Its study leads to several extensions of the classical inequality of Grothendieck, to an improvement of a recent result of Kashin and Szarek, and to a solution of a problem of Megretski and of Charikar and Wirth. Most of the talk is based on a joint work with Noga Alon, Yury Makarychev and Assaf Naor. 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/20070329/856e5ef6/attachment-0001.htm