February 2011 - talks - lists.cs.princeton.edu

colloquium speaker Maneesh Agrawala Wed. March 2
by Nicole E. Wagenblast 28 Feb '11

28 Feb '11

Designing Maps to Help People Maneesh Agrawala, Berkeley Wednesday, March 2, 4:30pm Computer Science 105 Over the last 15 years spatial databases and mapping services have become one of the most important applications on the Internet. In that time we have seen transformative increases in the availability of spatial data such as detailed satellite, streetside and casual images, 3D terrain and building models, and real-time traffic information. Such data has made it possible to create a variety of new kinds of maps and online experiences of any location on earth. Yet, only a handful of different kinds of maps are available online today. Moreover, it is not always clear what task today's online maps are designed to solve. In this talk I'll present a user-centered view of map design. I'll describe some of the difficult spatial tasks people regularly face as they navigate the world and I'll draw on the long history of mapmaking to show a variety of maps that are carefully designed to address these tasks. Examples will include maps designed to help users find their way from one location to another, to aid viewers in better understanding the spatial layout of 3D environments, and to assist tourists in finding points of interest in a new city. I'll conclude with a set of open challenges for online map design. Maneesh Agrawala is an Associate Professor in Electrical Engineering and Computer Science at the University of California, Berkeley. He received a B.S. in 1994 and a Ph.D. in 2002, both from Stanford University. Maneesh works on visualization, computer graphics and human computer interaction. His focus is on investigating how cognitive design principles can be used to improve the effectiveness of visual displays. The goals of this work are to discover the design principles and then instantiate them in both interactive and automated design tools. Recent projects based on this methodology include automated map design algorithms, systems for designing cutaway and exploded view illustrations of complex 3D objects, and interactive tools for manipulating digital photographs and video. He received an Okawa Foundation Research Grant in 2006, an Alfred P. Sloan Foundation Fellowship, an NSF CAREER award in 2007, the SIGGRAPH Significant New Researcher award in 2008 and MacArthur Foundation Fellowship in 2009.

1 1

Colloquium Speaker Andrew Adams Mon. Feb. 28
by Nicole E. Wagenblast 25 Feb '11

25 Feb '11

Systems and Algorithms for Computational Photography Andrew Adams, Stanford University Monday, February 28 4:30pm Computer Science 105 Digital cameras have become one of the most ubiquitous forms of computer in existence, but we still don't have a good handle on how to program them, what data we should program them to capture, how that data should be combined to make the photograph, and if a photograph is even the desired output. I'll talk about two of the tools we've developed to explore these questions. The first is the "Frankencamera", which is an open API and architecture for programmable cameras that makes it easy to deterministically control the various components of a camera at high frame rates. The goal of the Frankencamera is to enable research in photography, and indeed in graduate-level classes we've found that students equipped with this API are much more able to prototype their novel camera applications than with conventional camera APIs. The second tool is the discrete Gauss transform, which represents a class of algorithms that are widely used in photographic image processing for denoising, sharpening, and controlling tone and contrast. Incarnations of the discrete Gauss transform include the bilateral filter, the joint bilateral filter, and non-local means. We'll talk about making discrete Gauss transforms fast by representing the problem as resampling in a higher-dimensional space. Finally, time permitting, I'll talk about what we need to do to get these kinds of photographic image processing algorithms to run efficiently on the agglomeration of processors found inside today's cameras. Andrew Adams is a PhD student working with Marc Levoy at Stanford University. His research interests include computational cameras, fast image processing, and more generally languages and APIs that simplify complex domains. Andrew received a BS in Computer Science and Mathematics from the University of New South Wales in Australia in 2004, and an MS in Computer Science from Stanford in 2006. He has also worked on the Google Streetview project, and collaborates closely with Nokia Research on computational photography. Andrew also enjoys teaching, and has won several teaching awards.

1 1

Y Zhu preFPO
by Melissa Lawson 23 Feb '11

23 Feb '11

Yaping Zhu will present her preFPO on Tuesday March 1 at 10:30AM in Room 402. The members of her committee are: Jen Rexford, advisor; Vivek Pai and Aman Shaikh (AT&T Labs Research), readers; Mike Freedman and Andrea LaPaugh, nonreaders. Everyone is invited to attend her talk. Her abstract follows below. -------------------------------------- Minimizing Wide-area Performance Disruptions in Inter-domain Routing Abstract The Internet is the platform for most of our communications needs today. However, network changes like routing changes or congestion lead to performance disruptions, which would affect the user experience. Therefore, minimizing performance disruptions is crucial, and network operators have to react and fix the problems. Diagnosing wide-area performance disruptions is challenging: first, each network has limited visibility into the root cause of performance disruptions, requiring network operators to collect and analyze measurements of routing and traffic data; second, there are so many potential factors which might lead to performance disruptions, and these factors are usually interdependent with each other. Thus, the network diagnosis is usually done in an ad-hoc manner, and there are no formalized ways to define metrics and classify the performance disruption according to the causes. The thesis conducts two case studies to diagnose wide-area performance disruptions from the perspectives of a large tier-1 ISP and a large CDN: i) From the ISP's perspective, we designed and implemented a system that tracks inter-domain route changes at scale and in real time. Our system could be used as the building block for many diagnosis scenarios for the ISPs. ii) From the CDN's perspective, we focus on diagnosing wide-area network changes which caused latency increases to access the services in the CDN. We proposed a method for automatically classifying large increases of latency, and evaluated our techniques on one month of measurement data to identify major sources of high latency for a large CDN. Stepping back from the protocol designer's perspective, we refactor the inter-domain routing protocol BGP (Border Gateway Protocol), based on the lessons learned from the case studies: first, since each network only has limited visibility and control within its own network and the neighbors, we propose to select a route only based on the next- hop AS. second, the BGP protocol is not designed with operational challenges of performance and security in mind. Thus, there are many proposals to add additional BGP attributes and satisfy the operational needs. These proposals make the protocol and configuration complicated and error-prone, and make it difficult for network operators to diagnose problems. Instead, we argue to separate the performance and security requirements out of the protocol. Our proposal of next-hop BGP simplifies the protocol, and has the benefits of fast convergence, incentive compatibility, and easier support for multi-path routing.

1 0

Colloquium Speaker Aleksander Madry- Feb 23
by Nicole E. Wagenblast 21 Feb '11

21 Feb '11

Electrical Flows and Laplacian Systems: A New Tool for Graph Algorithms Aleksander Madry, Massachusetts Institute of Technology Wednesday, February 23rd, 4:30pm Computer Science Small Auditorium, CS105 In recent years, the emergence of massive computing tasks that arise in context of web applications and networks has made the need for efficient graph algorithms more pressing than ever. In particular, it lead us to focus on reducing the running time of the algorithms to make them as fast as possible, even if it comes at a cost of reducing the quality of the returned solution. This motivates us to expand our algorithmic toolkit to include techniques capable of addressing this new challenge. In this talk, I will describe how treating a graph as a network of resistors and relating the combinatorial properties of the graph to the electrical properties of the resulting circuit provides us with a powerful new set of tools for the above pursuit. As an illustration of their applicability, I will use these ideas to develop a new technique for approximating the maximum flow in capacitated, undirected graphs that yields the asymptotically fastest-known algorithm for this problem. Aleksander is a PhD candidate in Computer Science at MIT, advised by Michel Goemans and Jonathan Kelner. His research focuses on algorithmic graph theory, i.e. design and analysis of very efficient (approximation) algorithms for fundamental graph problems. He also enjoys investigating topics in combinatorial optimization - especially the ones involving dealing with uncertainty.

1 1

Colloquium speaker, Nick Feamster, Wed. Feb. 16th
by Nicole E. Wagenblast 15 Feb '11

15 Feb '11

Challenges and Opportunities for Tomorrow's Networks Nick Feamster, Georgia Tech Wednesday, February 16 , 4:30pm Computer Science Building, Small Auditorium, 105 The Internet is the underlying fabric for nearly all of our day-to-day communication; it is an undeniable success. More than a billion people around the world now have Internet access, and that number is projected to at least double in the next 10 years. The Internet is seeing increasing penetration in various resource-challenged environments, both in this country and abroad. These new frontiers for networking offer exciting opportunities, but they also pose serious challenges. Internet attacks are proliferating. Networks are becoming increasingly complex and difficult to troubleshoot. More than 20 countries around the world censor at least some Internet content. Our personal and private data is falling into the hands of a few large content providers. The performance of networks in homes, and how humans interact with them, are both poorly understood. These coming challenges will require us to think differently about networking: a discipline that has focused almost exclusively on performance must begin thinking more about people. In this talk, I will take a retrospective view of some important networking innovations over the past 10 years, surveying some solutions we and others have developed to improve the Internet's security, manageability and availability. I will then take a look forward at new challenges, and argue that to solve the most pressing problems for networking today, we must draw on techniques from a broad range of areas, including policy, economics, programming languages and human-computer interaction. Nick Feamster is an assistant professor in the College of Computing at Georgia Tech. He received his Ph.D. in Computer science from MIT in 2005, and his S.B. and M.Eng. degrees in Electrical Engineering and Computer Science from MIT in 2000 and 2001, respectively. His research focuses on many aspects of computer networking and networked systems, including the design, measurement, and analysis of network routing protocols, network operations and security, and anonymous communication systems. In December 2008, he received the Presidential Early Career Award for Scientists and Engineers (PECASE) for his contributions to cybersecurity, notably spam filtering. His honors also include a Sloan Research Fellowship, the NSF CAREER award, Technology Review's TR35 award, the IBM Faculty Fellowship, and award papers at SIGCOMM 2006, the NSDI 2005 conference, Usenix Security 2002 and Usenix Security 2001.

1 1

Colloquium Speaker, Mark Braverman, Tuesday Feb 15
by Nicole E. Wagenblast 14 Feb '11

14 Feb '11

Average-case solutions to hard problems Mark Braverman , University of Toronto Tuesday, February 15, 2011- 4:30pm Computer Science, 105 Many important computational problems are affected by various computational barriers that make them intractable in the worst case. In spite of this, it is often possible to get around these barriers and produce useful, if not optimal solutions for most instances of the problem. I will use several examples from my work to illustrate how considering an average or a typical case of the problem can be used to produce useful algorithms and mechanisms. The first example deals with the problem of assigning medical graduates to residency programs in the presence of married couples. The problem of finding a stable matching between residency programs and doctors without couples is solved by the famous Gale-Shapley algorithm. In the presence of couples, such a matching may not exist, and it is NP-hard to determine whether it exists or not. Still, in a large random market we show how to find a good stable outcome with high probability. The second example deals with aggregating noisy comparison signals, such as sport competition results, into a ranking most consistent with the observed signals. The worst-case version of the problem, known as the Minimum Feedback Arcset problem is NP-hard. We give an efficient algorithm for the average-case version of the problem.

1 0

Physics Colloquium-"Computer vision and optics in the study of fine art"
by Andrew W. Appel 04 Feb '11

04 Feb '11

1 0

S Jafarpour preFPO
by Melissa Lawson 04 Feb '11

04 Feb '11

Sina Jafarpour will present his preFPO on Friday February 11 at 2PM in Room 402. The members of his committee are: Robert Calderbank; advisor, Rob Schapire and Ingrid Daubechies (Duke); readers, and David Blei and Amit Singer (PACM and Math); non-readers. Everyone is invited to attend his talk. His abstract follows below. --------------------------------------------- Title: Deterministic Compressed Sensing The central goal of compressed sensing is to capture attributes of a signal using very few measurements. The initial publications by Donoho and by Candes and Tao have been followed by applications to image compression, data streaming, medical signal processing, digital communication and many others. The emphasis has been on random sensing but the limitations of this framework include performance guarantees, storage requirements, and computational cost. This talk will describe two deterministic alternatives. The first is based on expander graphs. We show that by reformulating signal reconstruction as a zero-sum game we can efficiently recover any sparse vector. We also demonstrate resilience to Poisson noise. The second is based on error correcting codes. We show that deterministic sensing matrices based on second order Reed Muller codes optimize average case performance. We also describe a very simple algorithm, one-step thresholding, that succeeds where more sophisticated algorithms, developed in the context of random sensing, fail completely. The talk is based on joint work with Waheed Bajwa, Robert Calderbank, Volkan Cevher, Marco Duarte, Zachary Harmany, Stephen Howard, Maxim Raginsky, Rob Schapire, and Rebecca Willett.

1 0