- talks - lists.cs.princeton.edu

U Syed preFPO
by Melissa Lawson 13 Feb '09

13 Feb '09

Umar Syed will present his preFPO on Friday February 20 at 10:30AM in Room 302. The members of his committee are: Rob Schapire advisor; David Blei and Warren Powell (ORF), readers; Michael Littman (Rutgers) and Yael Niv (PSY), non-readers. Everyone is invited to attend his talk. His abstract follows below. --------------------------- Title: Algorithms for Hybrid Supervised/Reinforcement Learning Problems Abstract: Reinforcement and supervised learning are two different approaches to machine learning. For an illustration their differences, consider the task of learning to drive a car. A reinforcement learning algorithm learns how to drive be trying several different driving behaviors, and uses feedback from the environment to decided which behavior is best. The feedback is delivered in the form of rewards, which reflect the desirability of various events such as crashing, staying on the road, etc. By contrast, a supervised learning algorithm passively observes an expert driver, and attempts to imitate his driving behavior. In general, reinforcement learning algorithms are concerned with maximizing reward, and supervised learning algorithms are concerned with matching observed behavior. In this talk, I will discuss learning algorithms that combine aspects of both reinforcement and supervised learning, leveraging the strengths of each approach. Specifically, I will describe algorithms for (i) using observations from an expert to learn high-reward behavior in environments where the true rewards are only partially known; (ii) improving the imitation of an expert's behavior by imposing reward-based constraints on the space of possible behaviors; (iii) learning high-reward behavior in an environment where the rewards can abruptly, but predictably, change. I will present theoretical performance guarantees for each of our algorithms, which in some cases represent substantial improvements over guarantees for earlier methods. Our algorithms are motivated by problems in vehicle navigation, spoken dialog systems, and web search. I will present experimental results from some of these domains. Includes joint work with Rob Schapire, Michael Bowling (U Alberta), Jason Williams (AT&T), Nina Mishra (Microsoft), and Alex Slivkins (Microsoft).

1 0

J Boyd-Graber preFPO
by Melissa Lawson 13 Feb '09

13 Feb '09

Jordan Boyd-Graber will present his preFPO on Tuesday February 17 at 10:30 AM in Room 302. The members of his committee are: David Blei, advisor; Christiane Fellbaum and Ryan MacDonald (Google), readers; Rob Schapire and Dan Osherson (Psych), nonreaders. Everyone is invited to attend his talk. His abstract follows below. ----------------------------- Title: Linguistic Extensions to Topic Models Topic Models have been an active area of research in recent years, and models like latent semantic analysis, probabilistic latent semantic indexing, and latent Dirichlet allocation (LDA) have been successfully used for detecting opinions, finding similar images, and finding relevant documents given a query. However, such models make few assumptions about the input data: words are unrelated to each other and the words in a document are completely exchangeable (the so-called bag of words model). In this work, we present algorithms that enhance the document-level knowledge provided by LDA with richer linguistic assumptions. First, we allow topic models to use words arranged in a tree (such as the WordNet ontology) rather than a simple flat list and derive inference using MCMC for this model. One application that is made possible by this change is word sense disambiguation, which discovers the meanings of words (i.e. discriminating between "bank" the financial institution and the landform). We show that incorporating topics in this model improves disambiguation accuracy. Secondly, we present a model that incorporates local syntactic information into topic models, which allows the algorithm to find groups of words that are both globally thematically consistent and locally syntactically consistent. We use the product of experts model to combine document and syntactic information and derive variational inference procedures for this model. We show that this model predicts word usage better than previous models. Yours, Jordan

1 0

last reminder: CS Colloquium (TODAY): Stuart Geman]
by Fei-Fei Li 11 Feb '09

11 Feb '09

-------- Original Message -------- Dear all, Stuart Geman, James Manning Professor of Applied Mathematics from Brown University, will give a CS Colloquium talk at 4:15pm TODAY in the CS Small Auditorium. Prof. Geman has been a leader in statistics and computer vision. His recent projects span from computer vision to neuroscience to financial markets. Below please find his talk title and abstracts. Fei-Fei Google and the Vapnik-Chervonenkis Dimension Stuart Geman Brown University Google engineers routinely train query classifiers, for ranking advertisements or search results, on more words than any human being sees or hears in a lifetime. A human being who sees a meaningfully new image every second for one-hundred years will not see as many images as Google has in its libraries, all of which are available for training object detectors and image classifiers. Yet by human standards the state-of-the-art, in computer understanding of language and computer-generated image analysis, is primitive. What explains the gap? Why can’t learning theory tell us how to make machines that learn as efficiently as humans? Upper bounds on the number of training samples needed to learn a classifier as rich and competent as the human visual system can be derived using the Vapnik-Chervonenkis dimension, or the metric entropy, but these suggest that not only does Google need more examples, but all of evolution might fall short. I will make some proposals for efficient learning and offer some mathematics to support them. _______________________________________________ Ml-stat-talks mailing list Ml-stat-talks(a)lists.cs.princeton.edu https://lists.cs.princeton.edu/mailman/listinfo/ml-stat-talks -- --------------------------------------------- Li, Fei-Fei Ph.D. (publish under L. Fei-Fei) Assistant Professor Computer Science Dept. Princeton University 35 Olden St. Princeton, NJ 08540 Tel: (609)258-8130 Office: Room 411, CS Building Website: http://vision.cs.princeton.edu ---------------------------------------------

1 0

reminder: Tomorrow's CS Colloquium (Feb 11): Stuart Geman]
by Fei-Fei Li 10 Feb '09

10 Feb '09

-------- Original Message -------- Subject: [Ml-stat-talks] CS Colloquium (Feb 11): Stuart Geman Date: Sun, 01 Feb 2009 17:08:43 -0500 From: Fei-Fei Li <feifeili(a)CS.Princeton.EDU> CC: csugrad(a)CS.Princeton.EDU, ml-stat-talks(a)lists.cs.princeton.edu, csgrad(a)CS.Princeton.EDU, csfac(a)CS.Princeton.EDU, pixl-talks(a)lists.cs.princeton.edu References: <8697FFE6-7521-4BB6-A8E8-7C1A55376400(a)cs.princeton.edu> Dear all, Stuart Geman, James Manning Professor of Applied Mathematics from Brown University, will give a CS Colloquium talk at 4:15pm Wednesday Feb 11. Prof. Geman has been a leader in statistics and computer vision. His recent projects span from computer vision to neuroscience to financial markets. Below please find his talk title and abstracts. Fei-Fei Google and the Vapnik-Chervonenkis Dimension Stuart Geman Brown University Google engineers routinely train query classifiers, for ranking advertisements or search results, on more words than any human being sees or hears in a lifetime. A human being who sees a meaningfully new image every second for one-hundred years will not see as many images as Google has in its libraries, all of which are available for training object detectors and image classifiers. Yet by human standards the state-of-the-art, in computer understanding of language and computer-generated image analysis, is primitive. What explains the gap? Why can’t learning theory tell us how to make machines that learn as efficiently as humans? Upper bounds on the number of training samples needed to learn a classifier as rich and competent as the human visual system can be derived using the Vapnik-Chervonenkis dimension, or the metric entropy, but these suggest that not only does Google need more examples, but all of evolution might fall short. I will make some proposals for efficient learning and offer some mathematics to support them.

1 0

Speaker Wednesday - 2/11
by Ginny Hogan 09 Feb '09

09 Feb '09

Talk Wed. - Small Auditorium (Room 105) 4:15pm Host: Fei-Fei Li Stuart Geman, Brown University - "Google and the Vapnik-Chervonenkis Dimension"

1 0

Colloquium Speaker Today -
by Ginny Hogan 04 Feb '09

04 Feb '09

Andrew Ng, Stanford, will present his Talk "STAIR: The Stanford Artificial Intelligence Robot Project" at 4:15pm today COS 105 -

1 0

Talk Tomorrow - Computer Science
by Ginny Hogan 03 Feb '09

03 Feb '09

Wednesday, February 4, 2009, 4:15 P.M. Small Auditorium (Room 105) Andrew Ng, Stanford, will present his talk STAIR: The STanford Artificial Intelligence Robot Project

1 0

Computer Science Speaker Wed Feb 4th -
by Ginny Hogan 02 Feb '09

02 Feb '09

Andrew Ng, Stanford, will present his talk "STAIR: The STanford Artificial Intelligence Robot Project" On Wed, Feb 4th, Computer Science Room 105 Talk starts at 4:15pm

1 0

W Kim generals
by Melissa Lawson 15 Jan '09

15 Jan '09

Wonho Kim will present his research seminar/general exam on Thursday January 22 at 2PM in Room 402. The members of his committee are; Vivek Pai (advisor), Jennifer Rexford, and Mike Freedman. Everyone is invited to attend his talk, and those faculty wishing to remain for the oral exam following are welcome to do so. His abstract and reading list follow below. ---------------------------------- Abstract Interactive development of large-scale networked systems is hard to achieve. We believe one of the key technology enablers for interactive development environment will be fast and efficient file synchronization since many distributed systems assume every change is synchronized with the participating nodes. However, most existing tools mainly focus on bandwidth efficiency and do not optimize the total completion time, which reduces both user interactivity and productivity. In this work, we explore design spaces for interactive file synchronization in heterogeneous networked systems environment. Specifically, we develop and compare a number of file distribution policies for minimizing the synchronization latency. Our policies prioritize remote nodes and files based on information dynamically available on run-time, and decide how to divide workloads between the origin server and external data dissemination services such as Contents Distribution Networks (CDN). Through simulations and evaluations on PlanetLab, we show that even a slight change in the policy can make drastic difference in terms of the total completion time under various scenarios. Reading list [Papers] [1] J. R. Albrecht, C. Tuttle, A. C. Snoeren, and A. Vahdat. Loose synchronization for large-scale networked systems. In Proceedings of the USENIX Annual Technical Conference, May 2006. [2] K. P. Birman, M. Hayden, O. Ozkasap, Z. Xiao, M. Budiu, and Y. Minsky. Bimodal multicast. ACM Transactions on Computer Systems, pages 4188, May 1999. [3] A. Muthitacharoen, B. Chen, and D. Mazières. A low-bandwidth network file system. In Proceedings of the ACM Symposium on Operating Systems Principles (SOSP), October 2001. [4] K. Park and V. S. Pai. Scale and performance in the CoBlitz large-file distribution service. In Proceedings of the Symposium on Networked Systems Design and Implementation (NSDI), May 2006. [5] J. Albrecht, R. Braud, D. Dao, N. Topilski, C. Tuttle, A. C. Snoeren, and A. Vahdat. Remote control: Distributed application configuration,management, and visualization with plush. In USENIX Large Installation System Administration Conference (LISA), Nov. 2007. [6] M. Harchol-Balter, B. Schroeder, N. Bansal, and M. Agrawal. Size-based scheduling to improve web performance. ACM Transactions on Computer Systems (TOCS), May 2003. [7] D. K. C, A. Rodriguez, J. Albrecht, and A. Vahdat. Bullet: High bandwidth data dissemination using an overlay mesh. In Proceedings of the ACM Symposium on Operating Systems Principles (SOSP), Oct. 2003. [8] H. Pucha, M. Kaminsky, D. G. Andersen, and M. A. Kozuch. Adaptive file transfers for diverse environments. In Proceedings of the USENIX Annual Technical Conference, June 2008. [9] A. Tridgell. Efficient Algorithms for Sorting and Synchronization. PhD thesis, The Australian National University, Feb. 1999. [Text Book] L. Peterson, and B. Davie, Computer Networks: A Systems Approach, 4th Edition, Morgan Kaufmann Publishers, March 2007.

1 0

C Wang generals - 10AM (time correction)
by Melissa Lawson 15 Jan '09

15 Jan '09

Re-announcing talk at correct time of 10AM. Sorry for the confusion. Chong Wang will present his research seminar/general exam on Wednesday Jan 21 at 10AM in Room 402. The members of his committee are: David Blei (advisor), Fei-Fei Li, and Rob Schapire. Everyone is invited to attend his talk, and those faculty wishing to remain for the oral exam following are welcome to do so. His abstract and reading list follow below. -------------------------------------------- Learning topic models from multiple corpora Abstract In this talk, we consider the problem of learning topic models from multiple corpora. Examples of multi-corpora data include news articles from different times or different locations, and scientific papers from different conferences or different years. The majority of topic models, however, usually ignore this addition information. Simply combining multiple corpora into one large corpus or treating each corpus individually can't provide the ability to analyze the high-level relations among multiple corpora, e.g., are topics similar or different from time to time or from location to location? How are they related to each other? I will describe two new approaches for learning topic models from multiple corpora: continuous time dynamic topic model (cDTM) and Markov topic model(MTM). In cDTMs, documents from the same time point are considered as a corpus. The cDTM extends dynamic topic models (DTMs) by using Brownian motion to model the latent topics through the time line. We derive an efficient variational approximate inference algorithm that takes advantage of the sparsity of observations in text, a property that lets us easily handle many time points. Thus, cDTM is able to discover topic evolutions in a much finer time resolution. In MTMs, papers from the same conference are treated as a corpus. Then we apply Gaussian (Markov) random fields to model the correlations of different corpora. MTMs capture both the internal topic structure within each corpus and the relationships between topics across the corpora. In addition, we will show cDTMs and DTMs can be formulated as special cases of MTMs. Quantitative results and qualitative discoveries (interesting topic patterns) will also be presented. Books: 1) Pattern Recognition and Machine Learning, by Christopher M. Bishop, Springer, 2006 chapters: 1, 2, 3.1-3.5, 4, 5, 8, 9, 10, 11.1-11.3, 12.1-12.2, 13, 14. 2) An Introduction to Probabilistic Graphical Models (unpublished manuscript), by Michael I. Jordan, 2002 chapters: 2, 3, 4, 5, 8, 9, 10, 11, 12, 13, 14, 15 4) (Optional) Artificial Intelligence: A Modern Approach, by Stuart Russell and Peter Norvig, Prentice Hall Series in Artificial Intelligence, 2003 chapters: 3.1-3.5, 4.1-4.3, 13, 14, 15.1-15.5, 18.1-18.3, 20.1-20.5, 23.2-23.3 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) T. Griffiths and M. Steyvers. Finding scientific topics. Proc Natl Acad Sci USA, 101 Suppl 1:5228-5235, April 2004. 4) C. Andrieu, N. de Freitas, A. Doucet, and M. Jordan. An introduction to MCMC for machine learning. Machine Learning, 50(1):5-43, January 2003. 5) M. Jordan, Z. Ghahramani, T. Jaakkola, and L. Saul. An introduction to variational methods for graphical models. Machine Learning, 37(2):183-233, 1999. 6) Y. Teh, M. Jordan, M. Beal, and D. Blei. Hierarchical Dirichlet processes. Journal of the American Statistical Association, 2006. 7) L. Fei-Fei and P. Perona. A Bayesian hierarchical model for learning natural scene categories. In CVPR, 2005. 8) Havard Rue and Turid Follestad, Gaussian Markov Random field models with applications in spatial statistics, preprint, 2003 9) D. Blei and J. Lafferty. Correlated topic models. In NIPS, 2005. 10) R. Kalman. A new approach to linear filtering and prediction problems. Transaction of the AMSE: Journal of Basic Engineering, 82:35-45, 1960. 11) L. Ruschendorf. Convergence of the iterative proportional fitting procedure. The Annals of Statistics, 23(4):1160-1174, 1995. 12) X. Wang and A. McCallum. Topics over time: a non-Markov continuous-time model of topical trends. In KDD, 2006. _______________________________________________ talks mailing list talks(a)lists.cs.princeton.edu https://lists.cs.princeton.edu/mailman/listinfo/talks

1 0