SUNMONTUEWEDTHUFRISAT
Conferences, Lectures, & Seminars
Events for March
-
CS Colloq: Dr. Julia Stoyanovich
Mon, Mar 01, 2010 @ 03:30 PM - 05:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Talk Title: BARAC: An Effective Presentation of Ranked Structured Datasets
Speaker: Dr. Julia Stoyanovich, University of Pennsylvania
Host: Prof. Cyrus ShahabiAbstract:
In online applications such as Yahoo! Personals and Yahoo! Real Estate users define structured profiles in order to find potentially interesting matches. Typically, profiles are evaluated against large datasets and produce thousands of matches. In addition to filtering, users also specify ranking in their profile, and matches are returned in the form of a ranked list. Top results in a ranked list are often homogeneous, which hinders data exploration. For example, a user looking for 1- or 2-bedroom apartments sorted by price will see a large number of cheap 1- bedrooms in undesirable neighborhoods before seeing any apartments with different characteristics. An alternative to ranking is to group matches on common attribute values (e.g., cheap 1-bedrooms in good neighborhoods, 2-bedrooms with 2 baths, etc.). However, not all groups will be of interest to the user given his ranking criteria. We argue here that neither single-list ranking norattribute-based grouping is adequate f or effective exploration of ranked datasets. We formalize rank- aware clustering and develop BARAC, a novel clustering algorithm that enables rank-aware data exploration in domains with a large number of heterogeneous attributes. We present results of a large- scale user study that validate the effectiveness of our approach.
We extensively evaluate the performance of our algorithm over large datasets from Yahoo! Personals, a leading online dating site.
Joint work with Sihem Amer-Yahia.Bio:
Julia Stoyanovich is a Postdoctoral Researcher and a Computing Innovations Fellow at the University of Pennsylvania. Julia holds M.S. and Ph.D. degrees in Computer Science from Columbia University, and a B.S. in Computer Science and in Mathematics and Statistics from the University of Massachusetts at Amherst. After receiving her B.S. Julia went on to work for two start-ups and one real company in New York City, where she interacted with a variety of massive datasets.
Julia's industry experience convinced her that many practical data management challenges remain to be tackled, and that she does not like to wake up early in the morning, prompting her return to academia.
Julia's research focuses on improving search, ranking, and data exploration in semantically rich application domains. She is particularly excited about the challenges that arise in life sciences applications and in social information processing.
Location: Ronald Tutor Hall of Engineering (RTH) - 306
Audiences: Everyone Is Invited
Contact: CS Front Desk
-
CS DLS: Prof. Daniela Rus
Tue, Mar 02, 2010 @ 03:30 PM - 05:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Talk Title: Programmable Matter with Self-reconfiguring Robots
Speaker: Prof. Daniela Rus
Hosts: Prof. Gaurav Sukhatme and Prof. Maja MataricAbstract:
We wish to create programmable matter by using robot modules capable of self-reconfiguration: hundreds of small modules autonomously organize and reorganize as geometric structures to best fit the terrain on which the robot has to move, the shape of the object the robot has to manipulate, or the sensing needs of the given task. Self-reconfiguration leads to versatile robots that can support multiple modalities of locomotion, manipulation, and perception.This talk will discuss the challenges of creating programmable matter, ranging from designing hardware capable of self-reconfiguration, to developing distributed controllers and planners for such systems that are scalable, adaptive, and support real-time behavior. We will discuss a spectrum of mechanical and computational capabilities for such systems and detail some recent self-reconfiguring robots.Bio:
Daniela Rus is a Professor of Electrical Engineering and Computer Science, where she is associate director of MIT's Computer Science and Artificial Intelligence Lab (CSAIL) and co-directs the MIT Center for Robotics at CSAIL. Her research interests include distributed robotics and mobile computing and her application focus includes transportation, security, environmental modeling and monitoring, underwater exploration, and agriculture.Rus is notable for spear-heading research in programmable matter by developing the several self-configuring robots. In addition, she worked with her students to has designed, control, and field autonomous underwater robots, agricultural robots that herd cattle, low-cost, early warning sensors for disaster prevention in developing countries, and teams of autonomous aerial vehicles that can monitor adaptively large environments.Rus is the recipient of the NSF Career Award and an Alfred P. Sloan Foundation Fellow. She is a Class of 2002 MacArthur Fellow and a fellow of AAAI and IEEE. Before receiving her appointment at MIT, Rus was a professor in the Computer Science Department at Dartmouth, where she founded and directed two laboratories in robotics and mobile computing.Rus earned her PhD in Computer Science from Cornell University.Location: Seaver Science Library (SSL) - 150
Audiences: Everyone Is Invited
Contact: CS Front Desk
-
CS Colloq: David Sontag
Thu, Mar 04, 2010 @ 03:30 PM - 05:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Talk Title: Approximate Inference in Graphical Models using LP Relaxations
Speaker: David Sontag
Host: Prof. Craig KnoblockAbstract:
Graphical models such as Markov random fields have been successfully applied to a wide variety of fields, from computer vision and natural language processing, to computational biology. Exact probabilistic inference is generally intractable in complex models having many dependencies between the variables.In this talk, I will discuss recent work on using linear programming relaxations to perform approximate inference. By solving the LP relaxations in the dual, we obtain efficient message-passing algorithms that, when the relaxations are tight, can provably find the most likely (MAP) configuration.Our algorithms succeed at finding the MAP configuration in protein side-chain placement, protein design, and stereo vision problems. More broadly, this talk will highlight emerging connections between machine learning, polyhedral combinatorics, and combinatorial optimization.Bio:
David is a Ph.D. candidate in Computer Science at MIT. He received his Bachelor's degree in Computer Science from the University of California, Berkeley in 2005. His research focuses on theory and practical algorithms for learning and probabilistic inference in very large statistical models. His work has been awarded with an outstanding student paper award at NIPS in 2007 and a best paper award at UAI in 2008. He currently has the Google Fellowship in Machine Learning.
Location: Seaver Science Library (SSL) - 150
Audiences: Everyone Is Invited
Contact: CS Front Desk
-
CS Colloq: Manish Bhide, IBM Research (India)
Fri, Mar 05, 2010 @ 10:00 AM - 11:00 AM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Talk Title:
Part 1: IBM Research - India Overview
Part 2: Keyword Search over Dynamic Categorized InformationSpeaker: Manish Bhide, IBM Research (India)Abstract:
My talk will be in two parts. In the first part I will give an
overview of IBM Research - India where I will outline the kind of work we do, the job opportunities, internship options, etc. In the second part of the talk I will present one of my research works titled "Keyword Search on Dynamic Categorized Information".
The abstract of the technical talk is given below:
Consider an information repository whose content is categorized. A data item (in the repository) can belong to multiple categories and new data is continuously added to the system. In this talk, I will describe a system, CS*, which takes a keyword query and returns the relevant top-K categories.
In contrast, traditional keyword search returns the top-K documents (i.e., data items) relevant to a user query. The need to dynamically categorize new data and also update the meta-data required for fast responses to user queries poses interesting challenges. The brute force approach of updating the meta-data by comparing each new data item with all the categories is impractical due to (i) the large cost involved in finding the categories associated with a data item and (ii) the high rate of arrival of new data items. We show that a sampling based approach which provides statistical guarantees on the reported results is also impracticable. We hence develop the CS* approach whose effectiveness results from its ability to focus on a strategically chosen subset of categories on the one hand, and a subset of new data on the other. Given a query, CS* finds the top-K categories with high accuracy even in time-constrained situations.Bio:
Manish Bhide is a Research Staff Member at IBM Research - India. He joined IBM Research in 2002 after finishing his masters from IIT Bombay. He is currently pursuing a part time PhD from IIT Bombay (expected completion Dec-2010 ). His research interests are primarily in the area of Information management. At IBM Research he has worked on areas such as XML, policy based data management information management issues in cloud computing, etc.Location: Seaver Science Library (SSL) - 150
Audiences: Everyone Is Invited
Contact: CS Front Desk
-
CS Colloq: Eftychios Sifakis
Tue, Mar 09, 2010 @ 03:30 PM - 05:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Talk Title: Digital doubles and the synergistic role of scientific computing, biomechanics and computer animationSpeaker: Eftychios Sifakis (University of California Los Angeles – Walt Disney Animation Studios)Host: Prof. Gerard MedioniAbstract:
Digital doubles have not only evolved into prevalent elements of motion pictures and games, but are also finding an ever growing application base including medical diagnostics, surgical planning and design of vehicles and crafts. At the same time, current and developing applications demand improved photorealism, enhanced biomechanical accuracy, better subject-specificity and faster simulation algorithms. As these demands often outgrow the evolution of computer hardware, new algorithms for biomechanical modeling and simulation are necessary to ensure that upcoming computational platforms are utilized to the best of their capacity. Additionally, biomechanical simulation has provided a great opportunity for transformative advances in medical practice using virtual models of the human body for disease prevention and treatment. These emerging applications mandate an increased level of attention to the unique demands of subject-specificity and anatomical accuracy for clinical uses of biomechanical modeling and simulation. This talk will outline a number of techniques that were designed to facilitate modeling and simulation of digital doubles with high fidelity and efficiency. Finally, I will discuss the cross-cutting impact of such advances on character animation, scientific computing and virtual surgery. Bio:
Eftychios Sifakis is a post-doctoral researcher at the University of California Los Angeles (with a joint appointment in Mathematics and Computer Science). He completed B.Sc. degrees in Computer Science (2000) and Mathematics (2002) from the University of Crete, Greece, and a Ph.D. degree in Computer Science (2007) from Stanford University. His research focuses on scientific computing, physics based modeling and computer graphics. He is particularly interested in biomechanical modeling for applications such as character animation, medical simulations and virtual surgical environments. Eftychios is a research consultant with Walt Disney Animation Studios, and has previously held consulting positions at Intel Corporation and SimQuest LLC.
Location: Seaver Science Library (SSL) - 150
Audiences: Everyone Is Invited
Contact: CS Front Desk
-
CS Colloq: Eng. Research Japan
Wed, Mar 10, 2010 @ 01:30 PM - 04:30 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Special Mini Symposium on Osaka University International Programs --Research Experience in Japan from Science to Technology--========================================1:30pm : Free Snacks (Pizza, Beverages)2-3:00pm:Cognitive Developmental Robotics as a bridge between
neuroscience and developmental psychology
Prof. Dr. Minoru ASADA, Eng.Intelligent Robotics at Osaka University
Assoc. Prof.Dr. Yasushi MAE, System Eng. Sci.Biomechanical Modeling in Physiome:Integration of
Multiscale Mechanics from Cell to Organ
Prof. Dr. Shigeo WADA, Mech. Eng. Sci.3-3:15pm: Coffee Break3:15pm-4:15pm:Bio-inspired Network Technologies for New Generation Networks
Assoc. Prof. Dr. Naoki WAKAMIYA, Information Sci.Astrophysics
Prof. Dr. Hideaki TAKABE, Laser Inst.Support for International Students and Scholars
Prof. Dr. J. J. CASTRO, and Ms. K. MATSUMURA, Int'l Student CenterMembrane Stress Biotechnology
Prof. Dr. Ryoichi KUBOI, Eng. Sci.Host: Stefan Schaal (sschaal@usc.edu)
Location: Hedco Neurosciences Building (HNB) - 100
Audiences: Everyone Is Invited
Contact: CS Front Desk
-
CS Colloq: Kamalika Chaudhuri
Thu, Mar 11, 2010 @ 03:30 PM - 05:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Talk Title: Statistical Algorithms for Modern Datasets
Speaker: Kamalika Chaudhuri
Host: Prof. Gaurav SukhatmeAbstract:
In this talk, we address two issues that arise in learning in modern datasets. First, with the increase in electronic record-keeping, many datasets that learning algorithms work with relate to sensitive information about individuals. Thus the problem of privacy-preserving learning -- how to design learning algorithms that operate on the sensitive data of individuals while still guaranteeing the privacy of individuals in the training set -- has achieved great practical importance. In this talk, we address the problem of privacy-preserving classification, and we present an efficient classifier which is private in the differential privacy model of Dwork et al. Our classifier works in the ERM (empirical lossminimization) framework, and includes privacy preserving logistic regression and privacy preserving support vector machines. We show that our classifier is private, provide analytical bounds on the sample requirement of our classifier, and evaluate it on some real data. A second characteristic of modern datasets is that data is often available from multiple domains or views. For example, when clustering a document corpus such as Wikipedia, we have access to the contents of the documents and their link structure. In this talk, we address this problem of Multiview Clustering -- how to use information from multiple views to improve clustering performance. We present an algorithm for multiview clustering, provide analytical bounds on the performance of our algorithm under certain statistical assumptions, and finally evaluate our algorithm on some real data.Based on joint work with Sham Kakade (UPenn), Karen Livescu (TTI Chicago), Claire Monteleoni (CCLS Columbia), Anand Sarwate (ITA UCSD), and Karthik Sridharan (TTI Chicago).Bio:
Kamalika Chaudhuri received a Bachelor of Technology degree in Computer Science and Engineering in 2002 from Indian Institute of Technology, Kanpur, and a PhD in Computer Science from UC Berkeley in 2007. She is currently a postdoctoral researcher at the Computer Science and Engineering Department at UCSD. Kamalika's research is on the design and analysis of machine-learning algorithms and their applications. In particular, her interests lie in -- clustering, online learning, and privacy-preserving machine-learning, and the applications of machine-learning and algorithms to practical problems in other areas.
Location: Seaver Science Library (SSL) - 150
Audiences: Everyone Is Invited
Contact: CS Front Desk
-
CS Colloq: Dr. Yaniv Altshuler
Tue, Mar 16, 2010 @ 03:30 PM - 05:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Talk Title: Collaborative Search of Expanding Grid DomainsSpeaker: Dr. Yaniv AltshulerHost: Prof. Sven KoenigAbstract:In nature, "simple minded" animals such as ants, bees or birds cooperate to achieve common goals and exhibit amazing feats of collaborative work. It seems that these creatures are "programmed" to interact locally in such a way that the desired global behavior is likely to emerge even if some individuals of the colony die or fail to carry out their task for other reasons.
A similar approach may be considered for coordinating a group of agents without a central supervisor, using only local interactions between the agents. Such agents can be either physical (e.g. robots) or virtual. When this decentralized approach is used, much of the communication overhead (typical of centralized systems) is saved, the resources required by the agents can be fairly limited, and better modularity is achieved. A properly designed system should be readily scalable, achieving reliability through redundancy.
In this talk I will discuss a set of analytic results concerning groups of simple and limited agents that are required to collaboratively and efficiently cover dynamic or expanding domainsBio: Dr. Yaniv Altshuler had received his PhD in Computer Science from the Technion, Israel Institute of Technology. For the past year, Dr. Altshuler has been a post-doc researcher at the Deutsche Telekom Lab in Ben Gurion university, at which he investigated the field of collaborative security.Location: Seaver Science Library (SSL) - 150
Audiences: Everyone Is Invited
Contact: CS Front Desk
-
Annual Research Review
Tue, Mar 23, 2010 @ 08:00 AM - 05:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
http://www.cs.usc.edu/researchreview2010/The review is an all-day event that showcases current research in the Computer Science department at USC. It will feature short research talks and posters by USC Computer Science faculty, postdocs, and PhD students.
Location: Davidson Conference Center (DCC)
Audiences: Everyone Is Invited
Contact: CS Front Desk
-
Taming the Scale and Costs of (Really) Large Distributed Systems
Wed, Mar 24, 2010 @ 03:30 PM - 05:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Talk Title: Taming the Scale and Costs of (Really) Large Distributed SystemsSpeaker: Dr. Harsha V. MadhyasthaHost: Prof. Ramesh GovindanAbstract: Over the last decade, the penetration of broadband Internet access and the commoditization of server hardware have dramatically increased. These trends have resulted in planetary-scale distributed applications that span millions of end-hosts and data centers that house hundreds of thousands of servers. Such large scales make it hard to build and deploy applications. In this talk, I will present simple models of these complex environments that help significantly improve the performance and cost-effectiveness of application deployments.First, I will present iPlane, an information plane designed to serve as the source of path information for all applications on the Internet. iPlane continually measures the Internet from several hundred geographically distributed vantage points to maintain an up-to-date map of the Internet's structure. By applying a structural model of the Internet on the data it gathers, iPlane can accurately predict properties such as latency, loss rate, and bandwidth along the path between arbitrary end-hosts in the Internet thus eliminating the need for measurement by any application. Over 3.5 years of deployment, iPlane has been used at more than 40 institutions, including to improve Google's content distribution network.Second, I will talk about BICMIC, a model that automates the process of determining the cluster configuration best suited to any particular data center application. BICMIC combines abstract representations of the application being deployed and the resources that can be used to construct the cluster to identify how various cluster configuration decisions should be combined to make the deployment cost-effective. Examples of configuration decisions include under-utilization of storage devices, caching of data in SSDs or DRAM, use of low-power CPUs, and separation of storage and compute into separate server farms. Bio: Harsha V. Madhyastha is a postdoctoral scholar at the University of California San Diego. He previously received his Ph.D. and M.S. degrees from theUniversity of Washington and his B.Tech. degree from the Indian Institute of Technology Madras, all in Computer Science and Engineering. He has been a
recipient of the Best Paper Award at the ACM SIGCOMM Internet Measurement Conference. His research interests span all aspects of distributed and networked systems.
Location: James H. Zumberge Hall Of Science (ZHS) - 159
Audiences: Everyone Is Invited
Contact: CS Front Desk
-
CS Colloq: Bryan Parno
Thu, Mar 25, 2010 @ 03:30 PM - 05:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Talk Title: Secure Code Execution on General-Purpose Computers
Speaker: Bryan Parno
Host: Prof. Ramesh GovindanAbstract:
As businesses and individuals entrust more and more sensitive tasks (e.g., paying bills, shopping online, or accessing medical records) to computers, it becomes increasingly important to ensure this trust is warranted. However, users are understandably reluctant to abandon the low cost, high performance, and flexibility of today's general-purpose computers. Thus, one of the fundamental questions I consider is: How can secure code execution coexist with the untrustworthy mountain of buggy yet feature-rich software that is common on modern computers?
For example, how can we keep a user's keystrokes private if the operating system, the most privileged software on the computer, cannot be trusted to be free of vulnerabilities? This is made all the more difficult by the need to preserve the system's existing functionality and performance.In this talk, I will present two techniques I have developed to address the need for features and security. With the Flicker architecture, I showed that that these conflicting needs can both be satisfied by constructing an on-demand secure execution environment, using a combination of software techniques and recent commodity CPU enhancements. This provides a solid foundation for constructing secure systems that must coexist with standard software; the developer of a security-sensitive code module need only trust her own code, plus as few as 250 lines of Flicker code, for the secrecy and integrity of her code's execution.Flicker assumes that a small portion of the computer's hardware can be trusted, but an increasing number of computing tasks are outsourced to the "cloud", where the user has no such guarantees. To formalize this setting, I introduced the notion of verifiable computing and designed a protocol to provably and efficiently provide computational integrity for work done by an untrusted party. The protocol also provides provable secrecy for the inputs and outputs of the computation. In addition, my protocol provides asymptotically optimal performance (amortized over multiple inputs). This result shows that we can outsource arbitrary computations to untrusted workers, preserve the secrecy of the data, and efficiently verify that the computations were done correctly.Bio:
Bryan Parno is a PhD candidate in Electrical and Computer Engineering
(ECE) at Carnegie Mellon University. He earned his Masters in ECE at Carnegie Mellon University, and his Bachelors in Computer Science at Harvard University. His current work focuses on the foundations of trust on modern computers. His research interests include computer security, systems, networks, and applied cryptography. In his spare time, he enjoys photography and volunteering as an Emergency Medical technician.
Location: Seaver Science Library (SSL) - 150
Audiences: Everyone Is Invited
Contact: CS Front Desk
-
Math Colloq: Prof. Van Vu
Mon, Mar 29, 2010
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Prof. Van Vu (Rutgers University)Host: Center for Applied Mathematical Sciences ( http://www.usc.edu/dept/LAS/CAMS/Colloquia/3-29-2010.html )Talk Title:
Inverse Littlewood-Offord theory, Smooth Analysis and the Circular LawAbstract:
A corner stone of the theory of random matrices is Wigner's semi-circle law, obtained in the 1950s, which asserts that (after a proper normalization) the limiting distribution of the spectra of a random hermitian matrix with iid (upper diagonal) entries follows the semi-circle law. The non-hermitian case is the famous Circular Law Conjecture, which asserts that (after a proper
normalization) the limiting distribution of the spectra of a random matrix with iid entries is uniform in the unit circle.Despite several partial results (Ginibre-Mehta, Girko, Bai, Edelman, Gotze-Tykhomirov, Pan-Zhu etc) the conjecture remained open for more than 50 years. In 2008, T. Tao and I confirmed the conjecture in full generality. I am going to give an overview of this proof, which relies on rather surprising connections between various fields: combinatorics, probability and, particularly, theoretical computer science.
Location: Kaprielian Hall (KAP) - 414
Audiences: Everyone Is Invited
Contact: CS Front Desk
-
CS Colloq: Alexandra Kolla
Tue, Mar 30, 2010 @ 03:30 PM - 05:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Talk Title: New Techniques for Using and Approximating Graph SpectraSpeaker: Dr. Alexandra Kolla Host: Prof. David KempeAbstract:
In this talk, we present novel techniques, based on spectral graph theory, and how they are used to design efficient algorithms for both practical and theoretical problems.
In the first part of the talk, we present new techniques for approximating a large graph with a smaller one. Namely, we show how, given a large graph G and a subgraph H of it, we can choose a very small number of edges H' out of H such that replacing H with H' does not change the spectrum of G by much.
We discuss significant implications of our techniques in two interesting practical problems: creating cost-efficient, well-connected networks and speeding up linear system solvers.
In the second part of the talk we present how spectral techniques can be useful for investigating the validity of Khot's Unique Games conjecture (UGC). UGC is one of the most central open problems in computational complexity theory. It asserts that for a certain constraint satisfaction problem, it is NP-hard to decide whether there is a labeling that satisfies almost all the constraints or, for every labeling, the fraction of the constraints satisfied is very small.
Since its origin, the UGC has been applied with remarkable success to prove tight hardness of approximation results for several important NP-hard problems such as Vertex Cover, MaxCut.
We discuss a novel spectral algorithm for deciding satisfiability of Unique Games. We show that our spectral approach works well on instances that previous techniques -which were solely based on linear and semidefinite
programming- provably fail.Bio:
I got my PhD at U.C Berkeley. My advisor was Umesh Vazirani. I am a postdoc at the Institute for Advanced Study in Princeton. My main area of interest is Theory of Computer Science and in particular spectral graph theory, convex programming and their implications to efficient algorithmic design. I am also interested in quantum computing and cryptography.
Bio: TBA
Location: Seaver Science Library (SSL) - 150
Audiences: Everyone Is Invited
Contact: CS Front Desk
