January 2010 - talks - lists.cs.princeton.edu

Colloquium Speaker, Wednesday, February 3, 2010-Tandy Warnow
by Michele J. Brown 28 Jan '10

28 Jan '10

Speaker: Tandy Warnow, University of Texas at Austin Title: *Simultaneous Alignment and Phylogenetic Tree Estimation* Wednesday, February 3, 2010 4:30 PM Small Auditorium (CS 105) Abstract: Molecular sequences evolve under processes that include substitutions, insertions, and deletions (jointly called \\\"indels\\\"), as well as other mechanisms (e.g., duplications and rearrangements). The inference of the evolutionary history of these sequences has thus been performed in two stages: the first estimates the alignment on the sequences, and the second estimates the tree given that alignment. While such methods seem to work well on relatively small datasets, these two-stage approaches can produce highly incorrect trees and alignments when applied to large datasets, or ones that evolve with many indels. In this talk, I will present a new method, SATe, that my lab has been developing that uses maximum likelihood to estimate the alignment and tree at the same time, and that can be used to analyze datasets with up to 1000 sequences on a desktop in 24 hours. Our study, using both real and simulated data, shows that this method produces much more accurate trees than the current best methods. Tandy Warnow is Professor of Computer Sciences at the University of Texas at Austin. Her research combines mathematics, computer science, and statistics to develop improved models and algorithms for estimating complex and large-scale evolutionary histories in both biology and historical linguistics. Tandy received her PhD in Mathematics at UC Berkeley under the direction of Gene Lawler, and did postdoctoral training with Simon Tavare and Michael Waterman at USC. She received the National Science Foundation Young Investigator Award in 1994, and the David and Lucile Packard Foundation Award in Science and Engineering in 1996. Tandy is a member of five graduate programs at the University of Texas, including Computer Science; Ecology, Evolution, and Behavior; Molecular and Cellular Biology; Mathematics; and Computational and Applied Mathematics. She is also the director for the multi-disciplinary CIPRES (Cyber-Infrastructure for Phylogenetic Research) Project, currently funded by the NSF under their Information Technology Program.

1 0

J Hung generals
by Melissa Lawson 15 Jan '10

15 Jan '10

Jack Tzu-Han Hung will present his research seminar/general exam on Wednesday January 20 at 9AM in room 402. The members of his committee are David August (advisor), Doug Clark, and Margaret Martonosi. Everyone is invited to attend his talk, and those faculty wishing to remain for the oral exam are welcome to do so. His abstract and reading list follow below. -------------------------------------------------------------- (Abstract) Recently, multicore systems have emerged as the dominant architecture in computing. Unfortunately, sequential programs do not benefit much from the new architecture. To keep up to the decades-old performance growth trend, some propose new programming languages or extensions for explicitly parallel programming [9, 2, 5, 3, 10, 8]. This approach leaves great burden to the programmers, including code rewriting effort and concerns for correctness and performance. An alternative is to use compilation techniques to extract the implicit parallelism from traditional, sequential programs. Most works focus on parallelizing the hottest loop in the program [12, 6, 4]; a few previous works attempt to exploit hybrid parallelism from several independent loops [11]. Hence, without considering the loop nest as a whole, these techniques leave a significant amount of parallelism unexploited. This work aims to extract various types of parallelism existing in a loop nest, using a dynamic programmingbased framework. Specifically, with a bottom-up traversal of the loop nest, every loop is examined to see if data or pipeline parallelism can be extracted and is assessed to see if the parallelization strategy is beneficial. The framework produces the final parallelized code that provides the optimal combination of the parallelization techniques. With manual parallelization, we experiment with six benchmarks, selected from different domains, including streaming applications (StreamIt benchmark) and data-intensive applications (SPECfp and MineBench). Our initial results show a (geo-mean) speedup of 3.35x (on an 8-core machine), while prior approaches (i.e., parallelizing one single loop or independent loops) only produce a speedup of 2.85x. (Reading List) [Papers] [1] J. R. B. Davies. Parallel loop constructs for multiprocessors. Master's thesis, Department of Computer Science, University of Illinois, Urbana, IL, May 1981. [2] M. I. Gordon, W. Thies, and S. Amarasinghe. Exploiting coarse-grained task, data, and pipeline parallelism in stream programs. In ASPLOS-XII: Proceedings of the 12th international conference on Architectural support for programming languages and operating systems, pages 151-162, New York, NY, USA, 2006. ACM. [3] J. Gummaraju, J. Coburn, Y. Turner, and M. Rosenblum. Streamware: programming general-purpose multicore processors using streams. In ASPLOS XIII: Proceedings of the 13th international conference on Architectural support for programming languages and operating systems, pages 297-307, New York, NY, USA, 2008. ACM. [4] J. Huang, A. Raman, Y. Zhang, T. B. Jablin, T.-H. Hung, and D. I. August. Decoupled software pipelining creates parallelization opportunities. In Proceedings of the 2010 International Symposium on Code Generation and Optimization, April 2010. [5] M. Kudlur and S. Mahlke. Orchestrating the execution of stream programs on multicore platforms. In PLDI '08: Proceedings of the 2008 ACM SIGPLAN conference on Programming language design and implementation, pages 114-124, New York, NY, USA, 2008. ACM. [6] G. Ottoni, R. Rangan, A. Stoler, and D. I. August. Automatic thread extraction with decoupled software pipelining. In Proceedings of the 38th Annual IEEE/ACM International Symposium on Microarchitecture, pages 105-116, November 2005. [7] E. Raman, G. Ottoni, A. Raman, M. Bridges, and D. I. August. Parallel-stage decoupled software pipelining. In Proceedings of the 2008 International Symposium on Code Generation and Optimization, April 2008. [8] W. Thies, V. Chandrasekhar, and S. Amarasinghe. A practical approach to exploiting coarse-grained pipeline parallelism in C programs. In Proceedings of the 40th Annual IEEE/ACM International Symposium on Microarchitecture, December 2007. [9] W. Thies, M. Karczmarek, and S. Amarasinghe. StreamIt: A language for streaming applications. In Proceedings of the 12th International Conference on Compiler Construction, 2002. [10] A. Udupa, R. Govindarajan, and M. J. Thazhuthaveetil. Synergistic execution of stream programs on multicores with accelerators. In LCTES '09: Proceedings of the 2009 ACM SIGPLAN/SIGBED conference on Languages, compilers, and tools for embedded systems, pages 99-108, New York, NY, USA, 2009. ACM. [11] H. Zhong, S. A. Lieberman, and S. A. Mahlke. Extending multicore architectures to exploit hybrid parallelism in singlethread applications. 2007. [12] H. Zhong, M. Mehrara, S. Lieberman, and S. Mahlke. Uncovering hidden loop level parallelism in sequential applications. In Proc. of the 14th International Symposium on High- Performance Computer Architecture, 2008. [13] L. Lamport. The parallel execution of DO loops. In Communications of the ACM, 1974. [14] J. Allen and K. Kennedy. Automatic loop interchange. In Proc. of the 1984 SIGPLAN symposium on Compiler construction, 1984. [15] W. Lee, R. Barua, M. Frank, D. Srikrishna, J. Babb, V. Sarkar, and S. Amarasinghe. Space-time scheduling of instruction-level parallelism on a raw machine. In Proc. of the eighth international conference on Architectural support for programming languages and operating systems, 1998. [16] A. Lim and M. Lam. Maximizing parallelism and minimizing synchronization with affine transforms. In the Proc. of the 24th ACM SIGPLAN-SIGACT symposium on Principles of programming languages, 1997. [17] A. Lim, G. Cheong, and M. Lam. An affine partitioning algorithm to maximize parallelism and minimize communication. In the Proc. of the 13th international conference on Supercomputing, 1999. [Text Books] [18] Modern Compiler Implementation in ML, by A. W. Appel, Cambridge University Press, 1998. [19] Compilers: Principles, Techniques, and Tools (2nd Edition) by Alfred V. Aho, Monica S. Lam, Ravi Sethi, and Jeffrey D. Ullman, Addison Wesley, 2006.

1 0

S Gerrish general exam
by Melissa Lawson 15 Jan '10

15 Jan '10

Sean Gerrish will present his research seminar/general exam on Wednesday January 20, at 11AM in room 402. The members of his committee are: David Blei (advisor), Ed Felten, and Christiane Fellbaum. Everyone is invited to attend his talk, and those faculty wishing to remain for the oral exam are welcome to do so. His abstract and reading list follow below. -------------------------------------- Modeling Influence in Text Corpora Identifying the most influential documents in a corpus is an important problem in many fields, ranging from information science and historiography to text summarization and news aggregation. A traditional method of assessing the impact of articles is to analyze the citations to it. The impact factor of a journal, for example, is based largely on academic citation analysis; and Googles successful PageRank algorithm is based on hyperlink citations between Webpages. Often, however, citation information is not present: certain legal documents, news stories, blog posts, sermons, and email, for example, all might lack such citation metadata, while there is a clear notion of influence among articles in these collections. In this talk, I will describe an unsupervised method for determining the influence of a document. Our intuition is that language changes over time, and that influential documents contribute to this change. I will formalize this intuition with a probabilistic graphical model and present an algorithm which takes a sequence of documents as input and computes a vector of "influence" for each document characterizing the document's influence. I will also discuss evaluation of this method and describe when this method provides a citation-free measure of bibliometric impact. Books / chapters 1. Selected chapters from Machine Learning and Pattern Recognition, Bishop Chapters: 1; 2; 3.1-3.2; 4.3-4.5; 7.1; 8; 9; 10; 11.1-11.4; 12.1-12.3; 14.1, 14.3 2. An Introduction to Probabilistic Graphical Models (unpublished manuscript), by Michael I. Jordan, 2002 Chapters: 2, 3, 4, 5, 8, 9,11 Papers: 1. D. Blei, A. Ng, and M. Jordan. "Latent Dirichlet Allocation", JMLR, 3:993-1022, 2003. 2. D. Blei and J. Lafferty. "Dynamic Topic Models", in ICML, 2006. 3. M. Jordan, Z. Ghahramani, T. Jakkola, and L. Saul. "An introduction to variational methods for graphical models.", Machine Learning, 37(2):183-233, 1999. 4. R. Nallapati and W. Cohen. "Link-plsa-lda: A new unsupervised model for topics and influence of blogs", in Proceedings of the International Conference on Weblogs and Social Media (ICWSM), 2008. 5. Gelman, A., Meng, X.L., & Stern, H. "Posterior predictive assessment of model fitness via realized discrepancies", Statistica Sinica, 6:733-807, 1996. 6. G. E. P. Box. "Sampling and Bayes' Inference in Scientific Modelling and Robustness", Journal of the Royal Statistical Society, 143(4):383-430. 1980. 7. L. Deitz, S. Bickel, and T. Scheffer. "Unsupervised prediction of citation influences", in Proceedings of the 24th International Conference on Machine Learning. Corvallis, Oregon, USA, June 2007. 8. D. Cohn and T. Hofmann. "The missing link - a probabilistic model of document content and hyptertext connectivity", in T. Leen et al, eds, Advances in Neural Information Processing Systems 13. 9. T. Griffiths and M. Steyvers. "Finding Scientific Topics", Proc Natl Acad Sci USA, 101 Suppl 1:5228-5235, April 2004. 10. A. Budanitsky and G. Hirst. "Evaluating WordNet-based Measures of Lexical Semantic Relatedness.", Computational Linguistics , 32 (1): 13-47, 2006. 11. Clinton, J. and Jackman, S. and Rivers, D. "The Statistical Analysis of Roll Call Data,", American Political Science Review. 98(2) 2004.

1 0

Xiaojuan Ma preFPO
by Melissa Lawson 06 Jan '10

06 Jan '10

Xiaojuan Ma will present her preFPO on Wednesday January 13 at 1:30PM in room 402. The members of her committee are: Perry Cook, advisor; Marilyn Tremaine (Rutgers) and Christiane Fellbaum, readers; Szymon Rusinkiewicz and Ken Steiglitz, nonreaders. Everyone is invited to attend her talk. Her abstract follows below. ------------------------------------------------- Title: A Multimedia Augmented Online Language Assistant for People with Aphasia and Other Language Barriers Abstract: Communication among language users take place usually in the spoken or written mode, as in face-to-face conversations, reading of books and newspapers, television watching and Internet browsing. However, for people with language disabilities (e.g. aphasia), people with low literacy, and people with poor command of a language, receiving and expressing information in this way is difficult. In particular, because of the inability to comprehend words and or to find words that express intended concepts, people with language barriers may be faced with great challenges performing everyday tasks such as ordering food in a restaurant and visiting a doctor. As an alternative to words, pictures and sounds have been designed, tested, verified and used to evoke concepts in computer interfaces, education, industry, and advertisement. However, icons created by artists, user-uploaded photos, auditory icons and earcons cannot always satisfy the need for communicating everyday concepts for people with language problems. We are building a Multimedia Augmented Online Language Assistant, which employs various multimedia forms, including web images, icons, animations, videos, and environmental sounds with the goal of assisting comprehension of common concepts. The Language Assistant is implemented as a popup dictionary in the form of a web browser extension. Users can select an unfamiliar word on a webpage to view its associated visual/auditory representation in a popup box. The Language Assistant can enhance concept understanding as people browse information on the Internet, and support face-to-face communication when people want to illustrate a term via a picture, a video or a sound when their conversation partner does not understand the spoken word. The Multimedia Augmented Online Language Assistant consists of a backend lexical network associated with large scale of multimedia data, and web interfaces to access the multimedia-concept associations. To develop the backend multimedia enhanced lexical database, we went through a cycle of design, construction, evaluation, and modification for each stimulus-to-concept associations. As an example we describe the creation of SoundNet, a nonspeech audio-to-concept semantic network which is part of the database. Soundnails, five-second environmental auditory representations, were extracted from sound effects libraries, and assigned to a vocabulary of common concepts. Two large scale studies were conducted via Amazon Mechanical Turk to evaluate the efficacy of soundnail representations. The first was a tagging study which collected human interpretations of the sources, locations, and interactions involved in the sound scene. The second study investigated how well people can understand words encoded in soundnails which are embedded in sentences compared to conventionally used icons and animations. Modification guidelines were proposed based on the study results. Through a series of studies, we verified that web images are as effective as stylized icons in conveying nouns, videos outperformed other stimuli in illustrating verbs, and nonspeech audio clips are better in distinguishing concepts like thunder, alarm, and sneezing. The Multimedia Augmented Online Language Assistant was shown to enhance information comprehension, and we are exploring its application in assisting communication involving medical care issues for people with language barriers and low levels of and medical literacy.

1 0