From macglashan at tti-c.org Fri Feb 15 11:45:54 2008 From: macglashan at tti-c.org (Julia MacGlashan) Date: Fri Feb 15 06:56:16 2008 Subject: [TTIC Colloquium] Abraham Flaxman, Carnegie Mellon University- TTI-C Talk Message-ID: <002101c86ffa$9f8a5790$aabf8780@jmacglDPLFYD1> When: Thursday, February 21, 2008 @ 10:00 am Where: TTI-C Conference Room Who: Abraham Flaxman, Carnegie Mellon University Topic: Trust-Based Recommendation Systems: An Axiomatic Approach High-quality, personalized recommendations are a key feature in many online systems. Since these systems often have explicit knowledge of social network structures, the recommendations may incorporate this information. This paper focuses on networks which represent trust and recommendations which incorporate trust relationships. The goal of a trust-based recommendation system is to generate personalized recommendation from known opinions and trust relationships. In analogy to prior work on voting and ranking systems, we use the axiomatic approach from the theory of social choice. We develop a natural set of five axioms which we desire any recommendation systems exhibit. Then we show that no system can simultaneously satisfy all these axioms. We also exhibit systems which satisfy any four of the five axioms. Next we consider ways of weakening the axioms, which can lead to a unique recommendation system based on random walks. We consider other recommendation systems (personal page rank, majority of majorities, and min cut) and search for alternative axiomatiztions which uniquely characterize these systems. Finally, we determine which of these systems are incentive compatible. This is an important property for systems deployed in a monetized environment: groups of agents interested in manipulating recommendations to make others share their opinion have nothing to gain from lying about their votes or their trust links. Contact: Sham Kakade, TTI-C sham@tti-c.org 773-834-2550 -------------- next part -------------- An HTML attachment was scrubbed... URL: http://ttic.uchicago.edu/pipermail/colloquium/attachments/20080215/cc011392/attachment.htm From macglashan at tti-c.org Mon Feb 18 16:19:31 2008 From: macglashan at tti-c.org (Julia MacGlashan) Date: Mon Feb 18 11:29:42 2008 Subject: [TTIC Colloquium] Abraham Flaxman, Carnegie Mellon University- TTI-C Talk Message-ID: <000001c8727c$55e2d7a0$aabf8780@jmacglDPLFYD1> When: Thursday, February 21, 2008 @ 10:00 am Where: TTI-C Conference Room Who: Abraham Flaxman, Carnegie Mellon University Topic: Trust-Based Recommendation Systems: An Axiomatic Approach High-quality, personalized recommendations are a key feature in many online systems. Since these systems often have explicit knowledge of social network structures, the recommendations may incorporate this information. This paper focuses on networks which represent trust and recommendations which incorporate trust relationships. The goal of a trust-based recommendation system is to generate personalized recommendation from known opinions and trust relationships. In analogy to prior work on voting and ranking systems, we use the axiomatic approach from the theory of social choice. We develop a natural set of five axioms which we desire any recommendation systems exhibit. Then we show that no system can simultaneously satisfy all these axioms. We also exhibit systems which satisfy any four of the five axioms. Next we consider ways of weakening the axioms, which can lead to a unique recommendation system based on random walks. We consider other recommendation systems (personal page rank, majority of majorities, and min cut) and search for alternative axiomatiztions which uniquely characterize these systems. Finally, we determine which of these systems are incentive compatible. This is an important property for systems deployed in a monetized environment: groups of agents interested in manipulating recommendations to make others share their opinion have nothing to gain from lying about their votes or their trust links. Contact: Sham Kakade, TTI-C sham@tti-c.org 773-834-2550 -------------- next part -------------- An HTML attachment was scrubbed... URL: http://ttic.uchicago.edu/pipermail/colloquium/attachments/20080218/97f2d2fc/attachment-0001.htm From macglashan at tti-c.org Thu Feb 21 14:54:07 2008 From: macglashan at tti-c.org (Julia MacGlashan) Date: Thu Feb 21 10:04:13 2008 Subject: [TTIC Colloquium] Jan Vondrak, Princeton University- TTI-C Talk Message-ID: <001c01c874cb$ec3c5860$aabf8780@jmacglDPLFYD1> When: Thursday, February 28, 2008 @ 10:00 am Where: TTI-C Conference Room Who: Jan Vondrak, Princeton University Topic: Approximation algorithms for combinatorial allocation problems Combinatorial allocation problems arise in situations where a set of items should be distributed among n players in order to maximize a certain social utility function. Such problems have been subject to recent interest due to their applications in combinatorial auctions and electronic commerce. Since allocation problems are typically NP-hard to solve optimally, we seek approximation algorithms that find a solution of value at least c * OPT where OPT is the optimum and c<1 a suitable constant. I will discuss the history of these problems and how they relate to classical work in combinatorial optimization. A case of particular interest is the Submodular Welfare Problem where utility functions are assumed to be monotone and submodular, a property known in economics as "diminishing returns". It has been known that a greedy algorithm yields a 1/2-approximation for this problem, and more generally for the problem of submodular maximization subject to a matroid constraint [Nemhauser, Wolsey, Fisher '78]. Among other results, I will show how this can be improved to a (1-1/e)-approximation - an approximation factor which is known to be optimal. A new ingredient in the algorithm is the approximate solution of a non-linear optimization problem using a "continuous greedy process". Contact: Julia Chuzhoy, TTI-C cjulia@tti-c.org 773-834-2490 -------------- next part -------------- An HTML attachment was scrubbed... URL: http://ttic.uchicago.edu/pipermail/colloquium/attachments/20080221/3d5b12ca/attachment.htm From macglashan at tti-c.org Wed Feb 27 09:17:41 2008 From: macglashan at tti-c.org (Julia MacGlashan) Date: Wed Feb 27 04:27:17 2008 Subject: [TTIC Colloquium] Jan Vondrak, Princeton University- TTI-C Talk Message-ID: <000301c87953$e5d5f530$aabf8780@jmacglDPLFYD1> When: Thursday, February 28, 2008 @ 10:00 am Where: TTI-C Conference Room Who: Jan Vondrak, Princeton University Topic: Approximation algorithms for combinatorial allocation problems Combinatorial allocation problems arise in situations where a set of items should be distributed among n players in order to maximize a certain social utility function. Such problems have been subject to recent interest due to their applications in combinatorial auctions and electronic commerce. Since allocation problems are typically NP-hard to solve optimally, we seek approximation algorithms that find a solution of value at least c * OPT where OPT is the optimum and c<1 a suitable constant. I will discuss the history of these problems and how they relate to classical work in combinatorial optimization. A case of particular interest is the Submodular Welfare Problem where utility functions are assumed to be monotone and submodular, a property known in economics as "diminishing returns". It has been known that a greedy algorithm yields a 1/2-approximation for this problem, and more generally for the problem of submodular maximization subject to a matroid constraint [Nemhauser, Wolsey, Fisher '78]. Among other results, I will show how this can be improved to a (1-1/e)-approximation - an approximation factor which is known to be optimal. A new ingredient in the algorithm is the approximate solution of a non-linear optimization problem using a "continuous greedy process". Contact: Julia Chuzhoy, TTI-C cjulia@tti-c.org 773-834-2490 -------------- next part -------------- An HTML attachment was scrubbed... URL: http://ttic.uchicago.edu/pipermail/colloquium/attachments/20080227/1d5204c9/attachment.htm From macglashan at tti-c.org Wed Feb 27 12:49:10 2008 From: macglashan at tti-c.org (Julia MacGlashan) Date: Wed Feb 27 07:58:43 2008 Subject: [TTIC Colloquium] Parikshit Gopalan, University of Washington- TTI-C Talk Message-ID: <002601c87971$70a18cc0$aabf8780@jmacglDPLFYD1> When: Wed, Mar 5, 2008 @ 10:00 am Where: TTI-C Conference Room Who: Parikshit Gopalan, University of Washington Topic: Fitting Polynomials to Noisy Data The problem of finding the polynomial that best fits a noisy data-set (or polynomial reconstruction) has a long history, dating back to curve-fitting problems studied in the 1800s. In the last two decades, there has been tremendous progress on this problem in computer science, driven by the discovery of powerful new algorithms. These results have spurred exciting new developments in Coding theory, Computational learning, Cryptography and Hardness of Approximation. In this talk, we will explore this problem from the perspectives of Coding theory and Computational learning. We begin with an algorithm for decoding a well-studied family of binary error-correcting codes called Reed-Muller codes, which are obtained from low-degree polynomials. The salient feature of this algorithm is that it works even when the number of errors far exceeds the so-called Johnson bound. I will present an algorithm for agnostically learning decision trees under the uniform distribution. This is the first polynomial time algorithm for learning decision trees in a harsh noise model. This algorithm solves the reconstruction problem for real polynomials using tools from convex optimization. I will also discuss settings where the reconstruction problem seems intractable. We will see evidence that the notorious Noisy Parity problem is hard under the uniform distribution. We will present hardness results suggesting that learning simple concepts with noise is impossible for arbitrary distributions. Contact: Julia Chuzhoy, TTI-C cjulia@tti-c.org 4-2490 -------------- next part -------------- An HTML attachment was scrubbed... URL: http://ttic.uchicago.edu/pipermail/colloquium/attachments/20080227/88baea48/attachment.htm