Conferences, Lectures, & Seminars
Events for October
-
CS Colloquium - Jitendra Padhye: AppInsight: Mobile App Performance Monitoring in the Wild
Wed, Oct 10, 2012 @ 12:30 PM - 02:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Jitendra Padhye, Microsoft Research
Talk Title: AppInsight: Mobile App Performance Monitoring in the Wild
Series: CS Colloquium
Abstract: The mobile-app marketplace is highly competitive. To maintain and improve the quality of their apps, developers need data about how their app is performing in the wild. The asynchronous, multi-threaded nature of mobile apps makes tracing difficult. The difficulties are compounded by the resource limitations inherent in the mobile platform. To address this challenge, we develop AppInsight, a system that instruments mobile-app binaries to automatically identify the critical path in user transactions, across asynchronous-call boundaries. AppInsight is lightweight, it does not require any input from the developer, and it does not require any changes to the OS. We used AppInsight to instrument 30 marketplace apps, and carried out a field trial with 30 users for over 4 months. We report on the characteristics of the critical paths that AppInsight found in this data. We also give real-world examples of how AppInsight helped developers improve the quality of their app.
Biography: Jiendra Padhye is a principal researcher at Microsoft Research. His primary research area is computer networking, with recent focus on data centers and mobile systems.
Host: Minlan Yu & Ethan Katz-Bassett
Location: Hughes Aircraft Electrical Engineering Center (EEB) - 248
Audiences: Everyone Is Invited
Contact: Assistant to CS chair
-
CS Colloquium: Anees Shaikh (IBM): Software-based Services for Cloud Networks
Wed, Oct 10, 2012 @ 03:00 PM - 04:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Anees Shaikh , IBM
Talk Title: Software-based Services for Cloud Networks
Series: CS Colloquium
Abstract: Software-defined networking (SDN) aims to provide a well-defined programming and automation interface to network devices. In moving network control and functionality to software running on standard server platforms, SDNs depart from the traditional, vertically integrated model of network hardware and software. We view SDNs as a new opportunity for supporting more seamless integration of networks with IT processes, and for providing higher-value network services.
In this talk, after describing the SDN model and our approach for building SDNs, we will focus on the problem of providing networking services for cloud computing platforms. As more enterprises look to leverage the cost and flexibility advantages of cloud computing, the lack of rich networking support remains a challenge. We will discuss the requirements of enterprise line-of-business applications for additional network functions in the cloud, and describe our research efforts to develop networking services for multi-tenant enterprise clouds using software-defined networking techniques.
Biography: Anees Shaikh is a Research Staff Member and Manager with the IBM TJ Watson Research Center in New York. He currently leads the Systems Networking Research group at Watson, focusing on problems related to data center and cloud networks, and works closely with the IBM Systems Networking development division on SDN and OpenFlow. Anees completed a Ph.D. in Computer Science and Engineering from the University of Michigan.
http://researcher.ibm.com/researcher/view.php?person=us-aashaikh
Host: Minlan Yu & Ethan Katz-Bassett
Location: Waite Phillips Hall Of Education (WPH) - B28
Audiences: Everyone Is Invited
Contact: Assistant to CS chair
-
CS Colloquium: Jun Zhu: Bayesian Inference with Max-margin Posterior Regularization
Thu, Oct 11, 2012 @ 03:30 PM - 05:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Jun Zhu, Tsinghua University
Talk Title: Bayesian Inference with Max-margin Posterior Regularization
Series: CS Colloquium
Abstract: Existing Bayesian models, especially nonparametric Bayesian methods, rely heavily on specially conceived priors to incorporate domain knowledge for discovering improved latent representations. While priors can affect posterior distributions through Bayes' theorem, imposing posterior regularization is arguably more direct and in some cases can be more natural and easier. In this talk, I will present regularized Bayesian inference (RegBayes), a computational framework to perform posterior inference with a convex regularization on the desired post-data posterior distributions. When the convex regularization is induced from a linear operator on the posterior distributions, RegBayes can be solved with convex analysis theory. Furthermore, I will present some concrete examples, including MedLDA for learning discriminative topic representations and infinite latent support vector machines for learning discriminative latent features for classification. All these models explore the large-margin idea in combination with a (nonparametric) Bayesian model for discovering predictive latent representations. I will discuss both variational and Monte Carlo methods for approximate inference.
Biography: Dr. Jun Zhu is an associate professor in the Department of Computer Science and Technology at Tsinghua University. His principal research interests lie in the development of statistical machine learning methods for solving scientific and engineering problems arising from artificial and biological learning, reasoning, and decision-making in the high-dimensional and dynamic worlds. Prof. Zhu received his Ph.D. in Computer Science from Tsinghua University, and his advisor was Prof. Bo Zhang. He did post-doctoral research with Prof. Eric P. Xing in the Machine Learning Department at Carnegie Mellon University. His current work involves both the foundations of statistical learning, including theory and algorithms for probabilistic latent variable models, sparse learning in high dimensions, Bayesian nonparametrics, and large-margin learning; and the application of statistical learning in social network analysis, data mining, and multi-media data analysis.
http://www.ml-thu.net/~jun/
Host: Fei Sha
More Info: http://www.cs.usc.edu/calendar/csevents.asp?date=10%2F11%2F2012
Location: Seaver Science Library (SSL) - 150
Audiences: Everyone Is Invited
Contact: Assistant to CS chair
Event Link: http://www.cs.usc.edu/calendar/csevents.asp?date=10%2F11%2F2012
-
CS Distinguished Lecture Series: Sanjeev Arora: Is Machine Learning Tractable? --- Three Vignettes
Tue, Oct 16, 2012 @ 03:30 PM - 05:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Sanjeev Arora, Princeton University
Talk Title: Is Machine Learning Tractable? --- Three Vignettes
Series: CS Distinguished Lectures
Abstract: Many tasks in machine learning (especially unsupervised learning) are provably intractable: NP-hard or worse. Nevertheless, researchers have developed heuristic algorithms to try to solve these tasks in practice. In most cases, these algorithms are heuristics with no provable guarantees on their running time or on the quality of solutions they return. Can we change this state of affairs?
This talk will suggest that the answer is yes, and describe three of our recent works as illustration. (a) A new algorithm for learning topic models. (It applies to Linear Dirichlet Allocations of Blei et al. and also to more general topic models. It provably works under some reasonable assumptions and in practice is up to 50 times faster than existing software like Mallet. It relies upon a new procedure for nonnegative matrix factorization.) (b) What classifiers are worth learning? (Can theory illuminate the contentious question of what binary classifier to learn: SVM, Decision tree, etc.?) (c) Provable ICA with unknown gaussian noise. (An algorithm to provably learn a "manifold" with small number of parameters but exponentially many "interesting regions.")
Biography: Sanjeev Arora is Charles C. Fitzmorris Professor of Computer Science at Princeton University. His research area spans several areas of theoretical Computer Science. He has received the ACM-EATCS Godel Prize (in 2001 and 2010), Packard Fellowship (1997), the ACM Infosys Prize for midcareer scientists (in 2012), the Fulkerson Prize (2012), the Simons Investigator Award (2012).
He served as the founding director for the Center for Computational Intractability at Princeton.
Host: Shaddin Dughmi
Location: Seaver Science Library (SSL) - 150
Audiences: Everyone Is Invited
Contact: Assistant to CS chair
-
CS Colloquium: Frank Dellaert: Subgraph Sparsifiers for Fast and Scalable Mapping and 3D Reconstruction
Thu, Oct 18, 2012 @ 03:30 PM - 05:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Frank Dellaert, Georgia Tech
Talk Title: Subgraph Sparsifiers for Fast and Scalable Mapping and 3D Reconstruction
Series: CS Colloquium
Abstract: Simultaneous Localization and Mapping (SLAM) and Structure from Motion (SFM) are important and closely related problems in robotics and vision. I will review how SLAM andSFM can be posed in terms of factor graphs, and that inference in these domains can be understood as variable elimination. After linearization, this corresponds to using direct linear solvers such as Cholesky and QR factorization. However, these methods face considerable challenges when confronted with graphs graphs that contain dense cliques, e.g., when seeing a tall tower from many different viewpoints. I will then show how identifying an efficient sub-problem (subgraph) can yield pre-conditioners for iterative methods to attack truly large-scale problems, and make the connection with support graph theory pioneered, among others, by Dan Spielman and Shang-Hua Teng.
Biography: Frank Dellaert is an Associate Professor in the School of Interactive Computing, College of Computing at Georgia Tech. His research is in the areas of Robotics and Computer vision. He is particularly interested in graphical model techniques to solve large-scale problems in mapping and 3D reconstruction. You can find out about his research and publications at http://www.cc.gatech.edu/~dellaert
Host: Gaurav Sukhatme
Location: Seaver Science Library (SSL) - 150
Audiences: Everyone Is Invited
Contact: Assistant to CS chair
-
Modeling HIV adaptation: Insights into HIV virology, immunology and vaccine design from machine learning and computational biology
Tue, Oct 23, 2012 @ 03:30 PM - 05:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Jonathan Carlson, Microsoft Research
Talk Title: Modeling HIV adaptation: Insights into HIV virology, immunology and vaccine design from machine learning and computational biology
Series: CS Colloquium
Abstract: The human immunodeficiency virus (HIV-1) mutates at a startling rate, with millions of viral variants generated in each patient each day. This high rate of mutation, coupled with high mutational tolerance, provides the virus with the ability to rapidly adapt to changing environments and typically proves an insurmountable challenge to the human immune system. Viral mutation is not, however, without constraints: given large enough datasets, patterns begin to emerge. By studying these patterns, we have gained significant new insights into what is attacking the virus (immunology), what is being attacked (virology), how that attack is evaded (evolution), and how adaptation influences disease progression (pathology). In addition, we have begun to identify features of individuals who naturally control the virus, offering tantalizing hints at how an effective vaccine might work. In this talk, I will describe the statistical models we have developed for studying HIV adaptation, the insights these models have provided and the open questions we continue to pursue.
Biography: Jonathan Carlson, Ph.D., joined the Escience Group at Microsoft Research in 2008, where he studies viral evolution, immunology and vaccine design through statistical modeling. His models of viral escape have achieved broad recognition in the HIV community, where they have led to the discovery of novel viral-host interactions, insights into mechanisms of natural immune control, and the identification of vaccine candidates that are slated for clinical trials. He has authored over 50 papers in the field and has served on advisory panels and committees for the Institutes of Medicine, the Gates Foundation and the Center for HIV/AIDS Vaccine Immunology (CHAVI). In 2009, he received his Ph.D. in computer science and computational molecular biology from the University of Washington, where he studied under David Heckerman (Microsoft Research) and Larry Ruzzo (UW) and was given the universityâs Distinguished Dissertation Award. He received his B.A. in Biology and Computer Science from Dartmouth in 2003, where he studied bioinformatics and transcriptional regulation under Bob Gross.
Host: Ethan Katz-Bassett
Location: Seaver Science Library (SSL) - 150
Audiences: Everyone Is Invited
Contact: Assistant to CS chair
-
CS Theory Lecture Series: Computing Divisor Class Groups of Function Fields Using Stark Units with Applications to Cryptography
Thu, Oct 25, 2012 @ 03:30 PM - 05:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Anand Narayanan, USC
Talk Title: Computing Divisor Class Groups of Function Fields Using Stark Units with Applications to Cryptography
Series: USC CS Theory
Abstract: Divisor class groups are structures central to the study of the arithmetic of global fields. We present a brief introduction to these groups and motivate their study in a computational setting by describing their extensive use in cryptography (elliptic and hyperelliptic curve based crypto-systems), error correction (algebraic geometric codes) and in solving certain Diophantine equations (Pell's Equation).
We then describe a new characterization of the structure of divisor class groups through the machinery of Kolyvagin systems from Stark units. This characterization leads to many interesting computational results; two of which we will discuss. The first is an efficient (nearly optimal) algorithm to compute the divisor class number of ray class fields. The second is a reduction relating the discrete logarithm problem in certain families of real elliptic/hyperelliptic curves to the principal ideal problem.
Host: Shang-Hua Teng
Location: SSL 150
Audiences: Everyone Is Invited
Contact: Assistant to CS chair
