2009-10 Dartmouth Computer Science Colloquium Schedule

NOTE NEW TIME: Seminars occur on Wednesdays from 4:15pm to 5:15pm in Moore 003 (unless otherwise noted), and are open to everyone. We are pleased to have you join us for light refreshments and tea at 3:45pm to meet the guest speaker and the audience.

Upcoming Seminars

Name: Joe Paradiso
Affiliation: Media Laboratory, MIT
Date: May 12, 2010
Host: Andrew Campbell and Tanzeem Choudhury
Title: Living with Ubiquitous Sensing and Dynamic Responsive Media
Abstract: This talk will overview recent work by Prof. Paradiso and his students at the MIT Media Lab that addresses the broad theme of interfacing humans to the ubiquitous electronic "nervous system" that sensor networks will soon extend across things, places, and people. He will show examples that involve cross-reality (everywhere blending of real and virtual worlds), personalization and security for ubiquitous media capture, wearable sensing, smart objects, human-computer interfaces and instrumented social interaction. He will also preview evolving work on interfaces to ubiquitous dynamic media that will engage the entire Media Lab as we begin to live in an environment with pervasive displays and sensors that track and derive context of occupants and deliver dynamic, personalized, and relevant content. Recent initiatives will be introduced that explore management of building utilities energy (e.g., heating, air conditioning, and lighting) via wearable and dispersed sensors, and recent results will be shown from an ongoing collaboration with the Boston Red Sox for monitoring and evaluating the performance of professional baseball players.
Bio: Joseph Paradiso is an Associate Professor of Media Arts and Sciences at the MIT Media Laboratory, where he directs the Responsive Environments group, which explores how sensor networks augment and mediate human experience, interaction, and perception. In addition, he co-directs the Things That Think Consortium, a group of industry sponsors and Media Lab researchers who explore the extreme fringe of embedded computation, communication, and sensing. After two years developing precision drift chambers at the Lab for High Energy Physics at ETH in Zurich, he joined the Draper Laboratory, where his research encompassed spacecraft control systems, image processing algorithms, underwater sonar, and precision alignment sensors for large high-energy physics detectors. He joined the Media Lab in 1994, where his current research interests include embedded sensing systems and sensor networks, wearable and body sensor networks, energy harvesting and power management for embedded sensors, ubiquitous and pervasive computing, localization systems, passive and RFID sensor architectures, human-computer interfaces, and interactive media. His honors include the 2000 Discover Magazine Award for Technological Innovation, and he has authored 200 articles and technical reports on topics ranging from computer music to power scavenging. After receiving a BS in electrical engineering and physics summa cum laude from Tufts University, Paradiso became a K.T. Compton fellow at the Lab for Nuclear Science at MIT, receiving his PhD in physics there for research conducted at CERN in Geneva.

Past Seminars (2009-2010)

Name: Lionel Levine
Affiliation: CSAIL, MIT
Date: October 7, 2009
Host: Peter Winkler
Title: Abelian Distributed Processors
Abstract: Abelian distributed processors are networks of interacting finite automata with a strong convergence property: the final output of the computation is the same regardless of the order in which the individual processors act. In some interesting cases, these networks obey a stronger property known as a "least action principle," which allows processors to act on messages they have not yet received. I'll use the abelian sandpile model to illustrate these ideas. In addition to the theorems, the talk will include pictures and simulations of the beautiful large-scale patterns formed by growing sandpiles on lattices.
Bio:

Name: Bill Freeman
Affiliation: CSAIL, MIT
Date: October 14, 2009
Host: Hany Farid
Title: Removing blur due to camera shake from images
Abstract: Blind deconvolution" is a beautiful, ill-posed problem: given an image that has been blurred by some unknown convolution kernel, estimate the image before it was it blurred. Lurking within this problem are nice, deep questions: What is an image, and how can you tell when one has been blurred? How should we solve very underdetermined inference problems?
We take a Bayesian approach, invoking a prior on both images and blur kernels. The choice of estimator is important for obtaining the desired solutions. An MAP estimate of the blur kernel and unblurred image typically favors interpreting the data as a sharp photo of a blurry subject, and thus favors not changing the blurry image. Instead, we can find the desired solutions by using an MMSE estimator, or, alternatively, by finding the MAP estimate but of only the blur kernel, marginalizing over possible image interpretations. I'll show our results on images people have submitted to us, and present our dataset for evaluating camera shake deconvolution algorithms.
The material of the talk is covered in these papers, and was joint work with the following authors:

R. Fergus, B. Singh, A. Hertzmann, S. Roweis, and W. T. Freeman.
Removing camera shake from a single image, SIGGRAPH 2006.
http://people.csail.mit.edu/billf/papers/deblur_fergus.pdf

A. Levin, Y. Weiss, F. Durand, and W. T. Freeman. Understanding and
evaluating blind deconvolution algorithms, CVPR 2009.
http://www.wisdom.weizmann.ac.il/~levina/papers/deconvLevinEtalCVPR09.pdf

Bio: Bill Freeman is a Professor of Electrical Engineering and Computer Science at MIT.

Name: Scott Aaronson
Affiliation: CSAIL, MIT
Date: October 21, 2009
Host: Amit Chakrabarti
Title: Quantum Money
Abstract: In a remarkable 1969 paper, Stephen Wiesner raised the possibility of money whose authenticity would be guaranteed by the laws of quantum physics, achieving something fundamentally impossible in the classical world. However, Wiesner's money can only be verified by the bank that printed it -- and the natural question of whether one can have secure quantum money that anyone can verify has remained open for forty years. In this talk, I'll tell you about recent progress on this question.

- I'll show that no "public-key" quantum money scheme can have security based on quantum physics alone: like in most cryptography, one needs a computational hardness assumption.

- I'll show that one can have quantum money that remains intractable to counterfeit, even if a counterfeiter gains black-box access to a device for checking the money.

- I'll describe an explicit candidate quantum money scheme I proposed last spring, and how that scheme was recently broken by Farhi, Gosset, Hassidim, Kelner, Lutomirski, Shor, and me.

- I'll describe a new quantum money scheme we propose in the same work. The new scheme has the strange property that not even the bank can prepare the same bill twice.

Reference for the first two results: S. Aaronson, "Quantum copy-protection and quantum money," in Proceedings of CCC'2009, http://www.scottaaronson.com/papers/noclone-ccc.pdf. The "AFGHKLS" paper is submitted and should be online soon.
Bio: Scott Aaronson is an Assisant Professor of Electrical Engineering and Computer Science at MIT.

Name: Calvin Newport
Affiliation:CSAIL, MIT
Date: October 28, 2009
Host: Dave Kotz
Title: Distributed Computing in the Age of Open Airwaves
Abstract: As we enter an era of ubiquitous computing, theoreticians are faced with a new scenario for distributed computation: large numbers of heterogeneous wireless devices competing to use the same unlicensed bands of the radio spectrum. The resulting conflicts ensure that unpredictable (sometimes even adversarial) interference is endemic throughout these bands. Traditional radio network models -- which typically assume a single predictable communication frequency used only by devices running the same algorithm -- do not adequately capture this reality. This motivates an important question: can theoreticians effectively model and prove useful bounds in this chaotic setting?

In this talk, I will propose the answer is "yes." Specifically, I will describe recent research in which my collaborators and I model an unpredictable radio band by incarnating the diversity of different interference behavior in an abstract adversary. This allows us to prove strong bounds that remain applicable to real world disrupted radio channels.To support this claim I will describe our research on the canonical problem of gossip, highlighting two interesting results in particular: first, there exist connections between oblivious deterministic solutions to gossip and extremal graph theory; and second, linear-time deterministic gossip is possible even in the presence of a moderate amount of adversarial interference.
Bio: I graduated Dartmouth it 2004. I then went to MIT where I earned my PhD in computer science in 2009. My advisor was Nancy Lynch and I specialized in distributed algorithms for wireless networks. I'm currently a postdoctoral associate working with Hari Balakrishnan on mobile mesh networks.

Name: Todd Zickler
Affiliation: EECS Harvard
Date: November 4, 2009
Host: Hany Farid
Title: Physics-based computer vision, reconsidered
Abstract: Computer vision systems are tasked with understanding an intricate visual world. A typical scene contains hundreds of surfaces that scatter light in distinct and complex ways, and these surfaces interact by occluding one another, casting shadows, and mutually reflecting light.

For computer vision systems to succeed they must find and exploit structure within this complexity, and this requires appropriate models of the way in which images are formed. Vision techniques that rely on such models are often labeled "physics-based", and traditionally, physics-based methods have achieved tractability by making rather severe assumptions about the world. They possess elegant formulations but are often hard to apply in natural settings.

Over the past few years, our group has been working to relax these restrictions by developing models that provide a better balance between tractability and accuracy. We seek computational models that are complex enough to accurately describe the visual world, and at the same time, are simple enough to be "inverted" for inference purposes. The hope is to build a stronger foundation for the computer vision systems of the future.

In this talk I will summarize our recent work and describe two approaches for recovering shape and reflectance information in natural environments.

Bio: Todd Zickler received a B.Eng. degree in honors electrical engineering from McGill University in 1996 and a Ph.D. degree in electrical engineering from Yale University in 2004. He subsequently joined the Harvard School of Engineering and Applied Sciences, where he is currently an associate professor. Todd's interests span computer vision, image processing, computer graphics, and human perception; and much of his work is devoted to developing efficient representations for appearance. He received an NSF career award in 2006 and was named an Alfred P. Sloan research fellow in 2008. More information can be found on his web-site: http://www.eecs.harvard.edu/~zickler.

Name: Andy Kern
Affiliation: Biology, Dartmouth College
Date: November 11, 2009
Host: Chris Bailey-Kellogg
Title: The tortoise and the hare: finding the targets of selection in genomes
Abstract: Among Kimura and Ohta's seminal “principles” of molecular evolution is the notion that functional
constraint determines rate of evolution at the genetic level. Indeed this principle lies at the intellectual
heart of the comparative genomics revolution of the past decade. I will demonstrate how
computationally sophisticated methods can be integrated with mathematical results from population
genetics and phylogenetics to identify those regions of genomes which are responsible for the myriad
innovations in biological evolution.
Bio: Andy Kern is an aasistant professor of Biological Sciences at Dartmouth

Name: James Aspnes
Affiliation: Computer Science, Yale University
Date: November 18, 2009
Host: Prasad Jayanti
Title: Max registers, counters, and monotone circuits
Abstract: Many shared-memory data structures act as mechanisms for combining the contributions of n different processes into a single value. Examples are counters, which return the sum of the contributions of the processes, and max registers, which return the maximum value. A classic lower bound of Jayanti, Tan, and Toueg shows that a non-blocking implementation of these data structures requires at least n-1 space and n-1 low-level read and write operations in the worst case. We show that this bound can be circumvented when the values in the data structure are bounded. For m-valued max registers, we give a wait-free, linearizable construction that implements read and write operations in only ceiling(log m) operations each, on an array of O(m) atomic bits. This construction is exactly optimal in the following sense: it is one of a family of closely-related algorithms parameterized by a choice of prefix-free code for the integers 0..m-1, and a matching lower bound shows that any sublinear algorithm for solving this problem is equivalent to some algorithm in this family.

Using the max-register construction, we show how to transform any monotone circuit into a wait-free concurrent data structure that provides write operations setting the inputs to the circuit and a read operation that returns the value of the circuit on the largest input values previously supplied. This gives an implementation of a wait-free, linearizable, polynomially-bounded counter with O(log n) cost per read and O(log2 n) cost per increment, an exponential improvement on the Omega(n) lower bound for unbounded counters. Joint work with Hagit Attiya and Keren Censor.
Bio:

Name: Rocco Servedio
Affiliation: Computer Science, Columbia University
Date: December 2, 2009
Host: Amit Chakrabarti
Title: Average Sensitivity of Polynomial Threshold Functions
Abstract: How many edges of the n-dimensional Boolean hypercube can be sliced by a degree-d polynomial surface? This question can be equivalently stated as "What is the maximum average sensitivity of any degree-d polynomial threshold function?" In 1994 Gotsman and Linial posed this question and gave a conjectured answer: the symmetric function slicing the middle d layers of the Boolean hypercube has the highest average sensitivity of all degree-d polynomial threshold functions.
In this work we give the first non-trivial upper bounds on average sensitivity of degree-d polynomial threshold functions (PTFs), thus making progress toward the Gotsman-Linial conjecture. The conjecture itself remains open.
I will explain the main ideas behind our result and describe some of its applications. These include
- the first polynomial-time agnostic learning algorithm for degree-d polynomial threshold functions;
- the first subexponential-time learning algorithm for AC0 circuits augmented with arbitrary linear threshold gates; and
- low-weight approximators for degree-d polynomial threshold functions.
Joint work with (various subsets of) Ilias Diakonikolas, Parikshit Gopalan, Prasad Raghavendra, Li-Yang Tan, and Andrew Wan.
Bio: Rocco Servedio is an Associate Professor of Computer Science at Columbia University in New York. He received his A.B. in Mathematics and
Ph.D. in Computer Science from Harvard, where he studied under Les Valiant. Rocco's research interests are centered around computational learning theory and computational complexity. Rocco has received an NSF Career Award and a Sloan Foundation Fellowship, and his research is currently supported by DARPA, Google, and the NSF.

Part of the special Get to know the CS department series of talks
Name: Devin Balkcom
Affiliation: Computer Science, Dartmouth
Date: January 6, 2010
Title: Robots and Geometry
Abstract: Even if we could build a robot with ten fingers, two hands, touch sensitive skin everywhere, and binocular vision, how would we write the algorithms to fold laundry, tie a knot, or even just move quickly from one place to another? We'll look at a collection of robotics problems for which the real challenge isn't building the robot, but doing the most with the least.

By the end of the talk, you'll know the fastest way to drive a planar mobile robot, be able to build a fingerless robot to tie knots, and be able to prove that a polygonal piece of cloth can be immobilized by grabbing no more than one-third of the convex vertices. I'll also show videos of our knot-tying and shirt-folding robots.
Bio: Devin Balcom is an Assistant Porfessor in the Computer Science department at Dartmouth College

Part of the special Get to know the CS department series of talks
Name: Emiliano Mulizzo and Nic Lane
Affiliation: Computer Science, Dartmouth
Date: January 13, 2010
Host: Andrew Campbell
Title: Mobile Phone Sensing is the Next Big Thing!
Abstract: New Apple, Google and Nokia smartphones that you use everyday come equipped with sensors (e.g., accelerometers, GPS, digital compasses, microphones, cameras, gyros) that enable new people-centric sensing applications across a wide variety of areas including social networks, personal environmental monitoring, distributed gaming, smart transportation systems and community healthcare assessment to name a few.

These smartphones are open, programmable and allow developers to distribute new sensing applications globally at Internet speed to millions of people. One can think of these phones as nodes in a global sensor network capable of not only personal sensing but also large-scale distributed sensing. This new technology push presents new and exciting opportunities and is an emerging area of research.

Many open questions remain to this vision. An open question driving our research is how much intelligence can be pushed to the phone to enable "smart" applications without undermining the main function of
the phone: to make calls, keep people connected, and browse the web. To make continuous sensing applications viable on mobile phones requires advances in sensing, inference and communications.

In this talk, we will discuss a number of projects that we have worked on in the Mobile Sensing Group (http://sensorlab.cs.dartmouth.edu/) including the CenceMe and SoundSense applications. These applications support lightweight machine learning techniques and low duty cycle operations and benefit from large-scale community sensor data to improve learning and classification.

SoundSense and community-guided learning is being developed jointly with Tanzeem Choudhury's group (http://pac.cs.dartmouth.edu/). CenceMe was developed in the ISTS MetroSense project.
Bio: Emiliano Miluzzo is a fifth year PhD candidate in the CS Department under the supervision of Prof. Andrew Campbell. He started his research working on problems related to static sensor network settings focusing on the design and development of new networking protocols and architectures. His current research focuses on the design, development, and evaluation of new frameworks, paradigms, and applications for people-centric mobile phone sensing systems. He recently completed an internship at Nokia Research Center, Palo Alto, working on novel mobile sensing mechanisms and applications.

Nicholas Lane is a 5th year graduate student at Dartmouth College working with Prof. Andrew Campbell. His research interests revolve around personal and community sensing systems built with mobile devices. He has worked on problems such as making robust and scalable inferences from the collected sensor data and dealing with resource and energy limitations. His work has been published in conferences such as SENSYS, MOBISYS and PERVASIVE. Nicholas recently returned from 5 months working at MSRA with Feng Zhao on mobile sensing systems.

Name: Subhadra Srinivasan
Affiliation: Thayer School of Engineering, Dartmouth
Date: January 20, 2010
Host: Tom Cormen
Title: Challenges in 3D Multi-modality Optical Imaging
Abstract: Multi-modality optical near infrared (MM-NIR) imaging combines traditional imaging techniques such as MRI, CT and X-Ray with optical methods to probe diseased tissues such as cancer non-invasively. The contrast in optical imaging is from higher absorption of light by malignant tissues due to higher hemoglobin concentration as compared to healthy and benign tissues. Combining this type of ‘functional’ information relating to metabolism with anatomical structure from X-Ray/MRI is attractive in the medical imaging community, with applications in breast cancer and brain imaging.

Finite element (FE) techniques provide a versatile solution to getting these images by solving a partial differential equation (PDE) for light transport with advantage of solving sparse matrix systems. In 3D the clinical workflow using FE is limited by need to create high-quality volume meshes of arbitrary shapes. Boundary element (BE) techniques offer an alternative by reducing the dimensionality of the problem to 2D assuming piece-wise constant regions in the tissue which are known a priori. While providing ease and reliability in meshing since only surfaces are required, the speed-up due to BE however depends on the problem due to dense matrix systems and reduces as the surface to volume ratio and number of regions increases. A hybrid BE-FE model has been recently developed that may provide a more optimal solution both in terms of speed and meshing. In the future GPU computing may provide a significant speed-up in these methods.

Even as numerical techniques evolve, the visualization of these MM-NIR datasets is limited by availability of software. To address this issue, we have recently developed a customized software platform using open source visualization toolkit (VTK) and Insight segmentation toolkit (ITK) for co-registration, display and analysis of optical estimates. All of these tools evolve to address the growing clinical need for fast and reliable data processing and innovative display tools.
Bio: Subha (pronounced Shuba) Srinivasan is Assistant Professor of Engineering at Thayer and new Adjunct Assistant Professor in CS. She completed her PhD in Biomedical Engineering from Dartmouth College in 2005. Her PhD thesis was based on improving quantification of multi-wavelength optical image reconstruction using different priors in 2D. Her post-doctoral work was in collaboration with Advanced Research Technologies, ART Inc. based in Montreal, Canada to optimize their clinical imager. In recent years, she has focused on 3D imaging and tools for increasing the computational efficiency especially in a multi-modality setting. Her work is funded through NIH grants. For more information, visit: http://engineering.dartmouth.edu/faculty/regular/subhadrasrinivasan.html

Name: Antonio Torralba
Affiliation: CSAIL, MIT
Date: January 27, 2010
Host: Lorenzo Torresani
Title: Understanding Visual Scenes
Abstract: Human visual scene understanding is remarkable: with only a brief glance at an image, an abundance of information is available - spatial structure, scene category and the identity of main objects in the scene. In traditional computer vision, scene and object recognition are two visual tasks generally studied separately. However, it is unclear whether it is possible to build robust systems for scene and object recognition, matching human performance, based only on local representations. Another key component of machine vision algorithms is the access to data that describe the content of images. As the field moves into integrated systems that try to recognize many object classes and learn about contextual relationships between objects, the lack of large annotated datasets hinders the fast development of robust solutions. In the early days, the first challenge a computer vision researcher would encounter would be the difficult task of digitizing a photograph. Even once a picture was in digital form, storing a large number of pictures (say six) consumed most of the available computational resources. In addition to the algorithmic advances required to solve object recognition, a key component to progress is access to data in order to train computational models for the different object classes. This situation has dramatically changed in the last decade, especially via the internet, which has given computer vision researchers access to billions of images and videos. In this talk I will describe recent work on visual scene understanding that try to build integrated models for scene and object recognition, emphasizing the power of large database of annotated images in computer vision.
Bio: Antonio Torralba is associate Professor of Electrical Engineering and Computer Science at the Computer Science and Artificial Intelligence Laboratory (CSAIL) at MIT. Following his degree in telecommunications engineering, obtained at the Universidad Politecnica de Catalunya, Spain, he was awarded a Ph.D. in Signal, Image, and Speech processing from the Institut National Polytechnique de Grenoble, France. Thereafter, he spent post-doctoral training at the Brain and Cognitive Science Department and the Computer Science and Artificial Intelligence Laboratory at MIT.

Name: Reza Olfati-Saber
Affiliation: Thayer School of Engineering, Dartmouth
Date: February 10, 2010
Host: Andrew Campbell
Title: Kalman-Consensus Filtering in Sensor Networks
Abstract:
Complex networked sensing and control systems with multi-modal sensing/actuation capabilities can be deployed for monitoring and reacting to the events occurring in an environment with a wide range of applications including intelligent transportation, large-scale human-in-the-loop systems, public health and environmental monitoring, and security and military applications.

One of the fundamental problems in sensor networks is to track the state of dynamic processes (targets) of interest using distributed algorithms that are scalable in the number of sensors and require local information exchange among neighboring nodes. In this talk, a novel distributed Kalman filtering algorithm called Kalman-Consensus Filter is introduced. This distributed estimation algorithm is capable of effectively tracking the state of single or multiple processes in static/mobile ad hoc wireless sensor networks. Formal results on convergence and performance properties of this distributed Kalman filtering algorithm will be provided. Furthermore, simulation results will be presented that illustrate the effectiveness of this network-wide cooperative filter.
Bio: Dr. Olfati-Saber is an Assistant Professor of Engineering at Dartmouth College. He received his SM and PhD degrees from MIT in Electrical Engineering and Computer Science in 1997 and 2001, respectively, and his BS degree from Sharif University of Technology. Prior to joining Dartmouth, he was a postdoctoral fellow at the Department of Control and Dynamical Systems at Caltech. He is the recipient of the 2008 NSF CAREER award and a member of IEEE, ASME, and Sigma Xi. His research interests include distributed estimation, control, and learning for cyber-physical networked systems, information fusion, and systematic design of self-organizing swarms

Name: Simon Kasif
Affiliation: Biomedical Engineering, Boston University
Date: February 17, 2010
Host: Chris Bailey-Kellog
Title: Can Computers Cure Cancer: From Genes and Genomes to Networks Signatures of Disease
Abstract: Biology textbooks suggest that genomes, from bacteria to human, contain the entire set of instructions which are theoretically sufficient for creating and executing molecular circuits that support the full repertoire of living organisms on earth. I will first review the progress and the challenges of a completely automated interpretation of the genetic code to infer what cells might be doing at different levels of accuracy. Some basic tasks we very very well, for instance, predicting where the genes are in any given genome. Other tasks, such as predicting the substrates (targets of enzymatic activity) for protein, we typically do not predict accurately. It is even more difficult to test such predictions systematically. During the talk we will review several fundamental predictive biology problems and describe previous successes, remaining challenges and current frameworks for solving these problems. The problem becomes more complex when we attempt to predict the behavior of a biological system in a given condition in general and disease in particular. I would outline some progress we made recently using network ideas to make and validate biologically and clinically relevant predictions in both cancer and diabetes.
Bio: Simon Kasif is a Professor of Bioengineering, Bioinformatics and Computer Science and Co-Director, Center for Advanced Genomic Technology (CAGT) at Boston University & Children's Hospital Informatics Program Harvard-MIT Division of Health Sciences and Technology (CHIP@HST).

Name: Adam Kalai
Affiliation: Microsoft Research New England
Date: February 24, 2010
Host: Lisa Fleischer
Title: Cooperation and Competition in Strategic Games
Abstract: There is an inherent tension between "cooperation" and "competition" in strategic games, especially in the presence of private information. One way to circumvent this difficulty is to allow binding agreements and side payments. But which agreement should be made, and how can it be made incentive compatible? This talk will present a general solution and applications to (a) routing and (b) selling hot dogs, with private information. No prior exposure to game theory is assumed.

This is joint work with Ehud Kalai.
Bio: Adam Kalai is a researcher at Microsoft Research New England.

Name: Henry Kautz
Affiliation: Computer Science, University of Rochester
Date: March 3, 2009 (POSTPONED)
Host: Tanzeem Choudhury
Title: Behavior Recognition and Interaction for Cognitive Assistance
Abstract: The synthesis of research in artificial intelligence, pervasive computing, and human-computer interaction is giving us the means to create systems that recognize human activity from low-level sensor data, interpret that information in light of commonsense theories of behaviors, plans, and goals, and ultimately provide help to the users in a natural and contextually appropriate manner. This talk describes how this approach has been used to develop assistive technologies for persons with cognitive disabilities.
Bio: Henry Kautz is Chair of the Department of Computer Science at the University of Rochester. He has been a professor at the University of Washington and department head at AT&T Bell Laboratories. He is President-Elect of AAAI and a Fellow of the AAAS.

Name: Nikhil Bansal
Affiliation: IBM Research Watson
Date: March 10, 2010
Host: Amit Chakrabarti
Title: A Primal-Dual Approach for Online Algorithms
Abstract: Online algorithms deal with settings where data arrives over time and current decisions must be made without the knowledge of future input. Until recently, analysis of online algorithms was based mostly on clever but ad-hoc techniques: usually relying on some potential function that seemed to be pulled out of thin air.

In this talk, I will describe a primal-dual approach for systematically designing and analyzing online algorithms. In addition to showing how it unifies and simplifies various previous results, I will also survey the recent success of this approach in addressing various outstanding problems such as weighted paging, and the randomized k-server problem.

Based on various joint works with Niv Buchbinder and Joseph (Seffi) Naor.
Bio:

Name: Jim Rehg
Affiliation: Georgia Tech
Date: March 31, 2010
Host: Tanzeem Choudhury
Title: Temporal Causality for Visual Event Analysis
Abstract: A basic goal of video understanding is the organization of video data into sets of events with associated temporal dependencies. For example, a soccer goal could be explained using a vocabulary of events such as passing, dribbling, tackling, etc. In describing the dependencies between events it is natural to invoke the concept of causality, but previous attempts to perform causal reasoning in video analysis have been limited to special cases, such as sporting events or naïve physics, where strong domain models are available. In this talk I will describe a novel, data-driven approach to the analysis of causality in video. The key to our approach is the representation of low-level visual events as the output of a multivariate point process, and the use of a nonparametric formulation of temporal causality to group event data into interacting subsets. This grouping process differs from standard motion segmentation methods in that it exploits the temporal structure in video over extended time scales. We apply our method to the analysis of complex social interactions in video. Specifically, we address the categorization and retrieval of social games between parents and children from unstructured video collections. This application is part of a larger effort in using computer vision technologies to support the detection, treatment, and understanding of developmental disorders such as autism.
Bio: Jim Rehg is a Professor in the School of Interactive Computing at the Georgia Institute of Technology. He is co-Director of the Computational Perception Lab and Associate Director of Research in the Center for Robotics and Intelligent Machines. He received his Ph.D. from CMU in 1995 and worked at the Cambridge Research Lab of DEC (and then Compaq) from 1995-2001, where he managed the computer vision research group. His research interests include computer vision, robotics, machine learning, and computer graphics. He recently served as the general co-chair for CVPR 2009 in Miami.

Name: Kevin Fu
Affiliation: Umass Amherst
Date: April 7, 2010
Host: Dave Kotz and Sean Smith
Title: Implantable Medical Devices: Security and Privacy for Pervasive, Wireless Healthcare
Abstract: An incredible array of implantable medical devices treat chronic ailments such as cardiac arrhythmia, diabetes, Parkinson's disease, seizures, and even obesity with various combinations of electrical therapy and drug infusion. These devices use tiny embedded computers to control therapies and collect physiological data. To improve patient care and detect early warning signs, implantable medical devices are rapidly embracing wireless communication and Internet connectivity. Implantable cardioverter defibrillators (ICDs) are wirelessly reprogrammable and relay medical telemetry over the Internet via at-home monitors. Such devices will vastly improve care for chronic disease, but will also introduce fundamentally new risks because of global computing infrastructures such as the Internet that are physically infeasible to secure. Thus, new devices must not only prevent accidental malfunctions, but must also prevent *intentional* malfunctions caused by malicious parties lurking on the network.

Our interdisciplinary research team implemented several software radio-based methods that could compromise patient safety and patient privacy (e.g., disclosing patient data or inducing ventricular fibrillation via a wireless command). Addressing these new risks, our zero-power approaches help to mitigate the risk of intentional malfunctions.

This work is joint with researchers from the University of Massachusetts Amherst, the University of Washington, and the Harvard Medical School.
Bio: Kevin Fu is an assistant professor in the Department of Computer Science at the University of Massachusetts Amherst in the beautiful northeastern region of the United States. Prof. Fu investigates how to ensure the security and privacy of pervasive devices that must withstand determined, malicious parties. To this end, Prof. Fu works on energy-aware cryptography and compiler techniques to run secure software on computational RFIDs---tiny embedded computers that operate without batteries. A computational RFID serves as a modular platform to test improvements in the security and privacy of medical devices. Prof. Fu's contributions include the security analysis of several systems ranging from contactless no-swipe credit cards and implantable cardiac defibrillators to access-controlled Web sites and automated software updates.

Kevin is an Alfred P. Sloan Research Fellow, MIT Technology Review TR35 Innovator of the Year, and recipient of the NSF CAREER award. His research appears in computer science conferences, medical journals, and has been featured in media such as The New York Times, The Wall Street Journal, and various news programs. He served on numerous program committees of leading conferences in secure systems, and has given dozens of invited talks world-wide to industry, government, and academia.

Prof. Fu leads the UMass Amherst Security and Privacy Research (SPQR) lab. He serves as director of the RFID Consortium on Security and Privacy (RFID-CUSP.org) and co-director of the Medical Device Security Center. Prof. Fu received his Ph.D. in Electrical Engineering and Computer Science from the Massachusetts Institute of Technology. He also holds a certificate of achievement in artisanal bread making from the French Culinary Institute and maintains an active participation in the study of Latin and the Classics. For more information, visit http://www.cs.umass.edu/~kevinfu/

Name: Michael Schapira
Affiliation:Yale University and UC Berkeley
Date: April 14, 2010
Host: Lisa Fleischer
Title: Internet Routing: Stability, Security and Selfishness
Abstract: The Internet is made up of smaller networks. The Border Gateway Protocol(BGP) establishes routes between these subnetworks, and is thus the glue that holds the Internet together. Over the past two decades there has been exponential growth in the scale and complexity of the Internet. However, BGP has not changed significantly in comparison and, consequently, does not always cope well with modern-day challenges. I shall present conceptual frameworks for addressing today's challenges and novel routing schemes that improve on the existing ones. Specifically, I shall present (1) a necessary condition for BGP safety, i.e., guaranteed BGP convergence to a ``stable'' routing outcome; (2) an economic approach to BGP security; and (3) a modest extension to BGP that enables more expressive routing policies while improving global network stability.
Bio:

Name: Jinbo Xu
Affiliation: Toyota Technical Institute, Chicago
Date: April 21, 2010
Host: Chris Bailey-Kellogg
Title: Probabilistic Graphical Model for Protein Structure Prediction
Abstract: If we know the primary sequence of a protein, can we predict its three-dimensional structure by computational methods? This is one of the most important and difficult problems in computational molecular biology and has tremendous implications for protein functional study and drug discovery.

Roughly speaking, existing computational methods for protein structure prediction can be broadly classified into two categories: template-based modeling (i.e, protein threading/homology modeling) and template-free modeling (i..e, ab initio folding). Template-based modeling predicts structure of a protein using experimental structure in the Protein Data Bank (PDB) as a template while template-free modeling predicts protein structure without depending on a template.

This talk will present new probabilistic graphical models for knowledge-based protein structure prediction. In particular, this talk will present a regression-tree-based Conditional Random Fields (CRF) method for protein threading and a CRF method for fragment-free approach to template-free modeling. Experimental results indicate that the CRF-based protein threading method outperforms all the other similar methods, especially on low-homology proteins. The CRF-based template-free modeling method is the first fragment-free approach that compares favorably to the popular fragment assembly method.
Bio:

Name: Bobby Blumofe
Affiliation: Akamai
Date: April 28, 2010
Host: Tom Cormen
Title: Using Mathematics and Algorithms to Make the Internet Scale
Abstract: Founded less than 12 years ago based on algorithms research at MIT, Akamai today delivers content and accelerates applications for over 3,000 customers, including many of the most popular social networking, video, and e-commerce sites. All of this runs on a massively distributed computing system that spans the globe with over 60,000 servers, in almost 1,000 networks, 71 countries, and 1,800 locations. All of this is made possible via mathematics and algorithms embodied in millions of lines of code that have been developed over the company's lifetime. Though this talk will not cover the algorithms themselves, it will cover some of the challenges and motivate the need for such algorithms in making the system work. In general, this talk is about the challenges of scale, focusing most specifically on the challenges that we see going forward into the next few years. This talk will show why it is that an Internet overlay system, such as Akamai, that uses good mathematics and algorithms is necessary to working at such scale.
Bio: As Senior Vice-President of Networks and Operations, Robert Blumofe is responsible for the deployment and operation of Akamai's production infrastructure consisting of over 60,000 servers distributed worldwide in 1,800 locations, spread among 70 countries, directly connected to 1,000 networks. This production infrastructure supports all of Akamai's services, delivering content and applications for over 3,000 enterprise customers, and routinely running at over 3 Tb/s and 10,000,000 hits/s. Robert is also one of Akamai's Chief Architects and serves on the Architecture Board. Prior to Akamai, Robert was an Associate Professor of Computer Science at the University of Texas at Austin where his research focused on algorithms and systems for parallel and distributed computing. Robert has a Ph.D. in Computer Science from MIT and a Sc.B. from Brown University.

Name: Umesh Shankar
Affiliation: Google Research
Date: May 5, 2010
Host: Dave Kotz
Title: Security and Data Integrity in Google Health
Abstract: Google Health is a Personal Health Record (PHR) application, allowing users to aggregate their medical data in one place, view it and make sense of it, combine it with relevant reference and scholarly information, and have control over who can access it. The sensitivity of the data requires great care to secure it and ensure users' privacy, and the diversity of data sources---users, medical providers, home medical devices, and mobile applications---requires flexibility in input methods. We will discuss the challenges that arise in authentication, authorization, and data provenance, as well as how we address these in practice. Both external protocols and a flexible, scalable model for crypto key management used internally will be covered. Last year's HITECH act renewed the push to make patient data available ubiquitously and electronically, but the problem of how to make a single coherent representation of it is unsolved. We will touch on a novel proposal for a very different method of representing medical records that addresses this issue.
Bio: Dr. Umesh Shankar is a security engineer at Google, focusing on user data protection and application security. He has published work in areas such as static analysis, network intrusion detection, ad-hoc networks, trusted computing, and usable security and privacy. He has also served on the program committees for numerous conferences. Dr. Shankar is currently a member of the SHARPS healthcare IT project's advisory board. He received a Ph. D. in computer science from the University of California, Berkeley, and an A.B. in computer science from Harvard University.

Past Seminars (2008-2009)

Name: Alex (Sandy) Pentland
Affiliation: Media Laboratory, MIT
Keynote Speaker for the Annual Computer Science Research Symposium
Date: Saturday, September 27, 2008
Time: 9:20 - 10:20am
Location: Rockefeller Center, Room 003
Host: James Hughes and Tanzeem Choudhury
Title: Honest Signals
Abstract: We are in the midst of an explosion of information about people and their behavior, but most of it is noise. Reality Mining sifts through this noise to discover the `honest signals' hidden within: subtle patterns that reliably reveal our relationships with other people, and accurately predict our future behavior. Honest signals offer an unmatched window into our financial, cultural, and organizational health. By understanding these subtle patterns we can better understand ourselves, and begin to create true collective intelligences.
Bio: Professor Alex (“Sandy”) Pentland is a pioneer in organizational engineering, mobile information systems, and computational social science. Sandy's focus is the development of human-centered technology, and the creation of ventures that take this technology into the real world. He directs the Digital Life Consortium, a group of more than twenty multinational corporations exploring new ways to innovate, and oversees the Next Billion Network, established to support aspiring entrepreneurs in emerging markets, and the EPROM entrepreneurship program in Africa. He is among the most-cited computer scientists in the world, and in 1997 Newsweek magazine named him one of the 100 Americans likely to shape this century

Name: Stephen Kobourov
Affiliation: AT&T Labs
Date: Wed, Oct 1, 2008
Time: 4:00 - 5:30pm
Host: Scot Drysdale
Title: Simultaneous Graph Embedding
Abstract: Traditional problems in graph visualization deal with a single graph while simultaneous graph visualization involves multiple related graphs. In the latter case nodes are placed in the same locations in all graphs and the graphs are simultaneously embeddable if crossing-free drawings for each graph can be found. We present polynomial time algorithms for simultaneous embedding of several classes of planar graphs and prove that some classes of graphs cannot be simultaneously embedded. More interesting are the half dozen variations: simultaneous geometric embedding, with and without mapping, colored simultaneous embeddings, near simultaneous embeddings, and matched embeddings.
Bio: Stephen Kobourov's research interests include information visualization, graph drawing and geometric algorithms, emphasizing problems relating to graph visualization. He received a BS in Computer Science and Mathematics from Dartmouth College in 1995 and PhD in Computer Science from Johns Hopkins University in 2000. He worked at the University of Arizona where he received a NSF Career grant and was tenured in 2006. As a Fulbright Scholar he spent a sabbatical year at the University of Botswana. He is an editor of the Journal of Graph Algorithms and Applications and has served on program committees for SODA, ESA, GD and SoftVis, and as program committee chair for the 10th Symposium on Graph Drawing. He joined AT&T Research Labs in 2008.

Name:Tamal Dey
Affiliation: Computer Science and Engineering, Ohio State University
Date: Wed, Oct 8, 2008
Time: 4:00 - 5:30pm
Host: Afra Zomorodian
Title: Delaunay Mesh Generation for Piecewise Smooth Domains
Abstract: Automatic mesh generation of various domains is an important goal in CAD. Recent sampling theory coupled with the Delaunay refinement strategy have proven to be effective for this problem. However, piecewise smooth surfaces and volumes remained as a challenge. Specifically, small angles subtended at non-smooth regions pose a serious obstacle to the strategy. In this talk we describe a method that removes the obstacle and can generate Delaunay meshes of piecewise smooth surfaces, volumes, and non-manifolds. The method is practical and has been implemented into a software called DelPSC which is available from http://www.cse.ohio-state.edu/~tamaldey/delpsc.html.
Bio: Tamal K. Dey is professor of computer science at The Ohio State University. His research interest includes computational geometry, computational topology and their applications in graphics and geometric modeling. After finishing his PhD from Purdue University in 1991 he spent a year in University of Illinois at Urbana Champaign as a post doctoral fellow. He has held academic positions in Indiana University-Purdue University at Indianapolis, IndianInstitute of Technology, Kharagpur, India, and Max-Planck Institute, Germany. Recently he authored a book ``Curve and surface reconstruction: Algorithms with mathematical analysis" published by Cambridge University Press. He leads the Jyamiti group which has developed various software including the well known Cocone software for surface reconstruction and DelPSC software for mesh generation. Details can be found at http://www.cse.ohio-state.edu/~tamaldey.

Name: Piotr Indyk
Affiliation: CSAIL, MIT
Date: Wed, Oct 15 2008
Time: 4:00 - 5:30pm
Host: Afra Zomorodian and Hany Farid
Title: Sparse Recovery Using Sparse Random Matrices
Abstract: Over the recent years, a new *linear* method for compressing high-dimensional data (e.g., images) has been discovered. For any high-dimensional vector x, its *sketch* is equal to Ax, where A is an m x n matrix (possibly chosen at random). Although typically the sketch length m is much smaller than the number of dimensions n, the sketch contains enough information to recover an *approximation* to x. At the same time, the linearity of the sketching method is very convenient for many applications, such as data stream computing and compressed sensing.
The major sketching approaches can be classified as either combinatorial (using sparse sketching matrices) or geometric (using dense sketching matrices). They achieve different trade-offs, notably between the compression rate and the running time. Thus, it is desirable to understand the connections between them, with the goal of obtaining the "best of both worlds" solution.
In this talk we show that, in a sense, the combinatorial and geometric approaches are based on different manifestations of the same phenomenon. This connection will enable us to obtain several novel algorithms and constructions, which inherit advantages of sparse matrices, such as lower sketching and recovery times.
Joint work with: Radu Berinde, Anna Gilbert, Howard Karloff, Milan Ruzic and Martin Strauss.
Bio: Piotr joined MIT in September 2000, after earning PhD from Stanford University. Earlier, he received Magister degree from Uniwersytet Warszawski in 1995. As of July 2007, he holds the title of Associate Professor with Tenure in the Department of Electrical Engineering and Computer Science. Piotr's research interests include: computational geometry (especially in high dimensional spaces), algorithms using sublinear time and/or space and streaming algorithms. He is also interested in algorithmic coding theory and pattern matching problems.

Name: Daniel Spielman
Affiliation: Computer Science, Yale University
Date: Wed, Oct 22, 2008
Time: 4:00 - 5:30pm
Host: Amit Chakrabarti
Title: Graph approximation and local clustering, with applications to the solution of diagonally-dominant systems of linear equations
Abstract: We discuss several fascinating concepts and algorithms in graph theory that arose in the design of a nearly-linear time algorithm for solving diagonally-dominant linear systems. We begin by defining a new notion of what it means to approximate a graph by another graph, and explain why these sparse approximations enable the fast solution of linear equations. To build these sparsifiers, we rely on low-stretch spanning trees, random matrix theory, spectral graph theory, and graph partitioning algorithms.
The need to quickly partition a graph leads us to the local clustering problem: given a vertex in a massive graph, output a small cluster near that vertex in time proportional to the size of the cluster.
We use all these tools to design nearly-linear time algorithms for solving diagonally-dominant systems of linear equations, and give some applications.
This talk focuses on joint work with Shang-Hua Teng, and touches on work by Vaidya, Gremban, Miller, Koutis, Emek, Elkin, Andersen, Chung, Lang, Daitch, Srivastava and Batson.

Bio:

Name: Josh Bongard
Affiliation: Computer Science, University of Vermont
Date: Wed, Oct 29, 2008
Time: 4:00 - 5:30pm
Host: Devin Balkcom
Title: Investigations at the Interface of Morphology, Evolution and Cognition
Abstract: Most computational attempts to understand cognition have focused on its proximate mechanisms: understanding the function of existing biological systems related to cognition and implementing them in artificial systems. In this talk I will discuss several robotics projects I have been involved in in which we try to shed light on the ultimate mechanisms of cognition: what selection pressures and task environments lead to the appearance of particular cognitive abilities, and what morphological and neural structures must evolve to support those abilities.
Bio: Josh Bongard received his Bachelors degree in Computer Science from McMaster University, Canada, his Masters degree from the University of Sussex, UK, and his PhD from the University of Zurich, Switzerland. He served as a postdoctoral associate under Hod Lipson in the Computational Synthesis Laboratory at Cornell University from 2003 to 2006. He is the co-author of the popular science book entitled "How the Body Shapes the Way We Think: A New View of Intelligence," MIT Press, November 2006 (with Rolf Pfeifer). Currently, he is an assistant professor in the Computer Science Department at the University of Vermont. His research interests include embodied cognition and evolutionary computation, and he was named both a Microsoft Research New Faculty Fellow in 2006, as well as a member of the TR35: MIT Technology Review's top 35 innovators under the age of 35.

Name: Rob Ghrist
Affiliation: U. of Pennsylvania
Date: Wed, Nov 5, 2008
Time: 4:00 - 5:30pm
Host: Afra Zomorodian
Title: Euler Calculus for Sensor Networks
Abstract: This work is motivated by a fundamental problem in sensor networks -- the need to aggregate redundant sensor data across a network. We focus on simple problems of enumerating and localizing targets using a network of minimal sensors that can detect and count nearby targets, but cannot identify or localize them. We solve this problem with calculus --- but not the calculus of Netwon et al. An integration theory built from algebraic topology and the Euler characteristic provides a computable, robust, and powerful tool for data aggregation.

Bio: Rob Ghrist is the Andrea Mitchell PIK Professor at the University of Pennsylvania, where he holds a joint appointment in the departments of Mathematics and Electrical and Systems Engineering. Before joining Penn, Ghrist was University Scholar and Richard and Margaret Romano Professional Scholar at the University of Illinois at Urbana-Champaign, where he held positions in the Department of Mathematics, Coordinated Science Laboratory, and Information Trust Institute. A winner of a Presidential Early Career Award for Scientists and Engineers (PECASE) and a Career Award from the National Science Foundation, Ghrist was named by Scientific American as a top 50 scientific innovator of 2007.

Name: Jason Hartline
Affiliation: Electrical Engineering and Computer Science, Northwestern University
Date: Wed, Nov 12, 2008
Time: 4:00 - 5:30pm
Location: Kemeny 007 (note different location this week)
Host: Lisa Fleischer
Title: Economics in Internet Design
Abstract: As the Internet has developed to become the single most important arena for resource sharing among parties with diverse and selfish interests, traditional algorithmic and distributed systems approaches are insufficient. To prevent undesirable Internet phenomena such as spam in email systems, bid-sniping in eBay's auction marketplace, free-loading in file-sharing networks, and click-fraud in Internet advertising; game-theoretic and economic considerations from auction theory must be applied. The first part of the talk is an overview of the foundational 2007 Economics Nobel prize winning theory of auction design. Unfortunately, this theory is insufficient for design of Internet algorithms both for its reliance on (micro) payments (whereas Internet protocols do not use payments) and because it suggests a different auction for every setting (whereas an Internet protocol must work well under a wide range of workloads). The second part of the talk develops an economic theory appropriate for Internet design.
Bio: Dr. Hartline joined the EECS department at Northwestern University in January of 2008. Prior to joining Northwestern, he spent four years as researcher at Microsoft Research, Silicon Valley, where he studied foundational topics in Internet economics and applied them, e.g., to ad auctions. He was a founding organizer of the Bay Algorithmic Game Theory Symposium. In 2003, he held a postdoctoral research fellowship at the Aladdin Center at Carnegie Mellon University and he received his Ph.D. in Computer Science from the University of Washington in 2003 advised by Anna Karlin. In third grade he got first place in a Halloween window painting contest. www.eecs.northwestern.edu/~hartline/

Name: Martha E. Pollack
Affiliation: Electrical Engineering and Computer Science and School of Information, U. of Michigan
Date: Wed, Nov 19, 2008
Time: 4:00 - 5:30pm
Host: Tanzeem Choudhury
Title: Intelligent Assistive Technology: The Present and the Future
Abstract: Recent advances in two areas of computer science—wireless sensor networks and AI inference strategies—have made it possible to envision, and to begin developing, a wide range of technologies that can improve the lives of people with physical, cognitive, and/or psycho-social impairments. This talk will focus on assistive technology for people with cognitive impairment, surveying the state-of-the-art and speculating about future design challenges and opportunities. A key theme will be that the usefulness and effectiveness of these systems depends on their being adaptive to the often highly individualized and changing needs of their users, and that to achieve these properties, designers must integrate a variety of strategies for automated reasoning and learning.
Bio: Martha E. Pollack is Dean and Professor in the School of Information at the University of Michigan, where she is also Professor of Computer Science and Engineering. She received her B.A. degree from Dartmouth College and her M.S.E. and Ph.D. degrees from the University of Pennsylvania, and has been a faculty member at the University of Pittsburgh and a research staff member at the AI Center at SRI International. A Fellow of the Association for the Advancement for Artificial Intelligence, for which she is also President-Elect, Dr. Pollack serves on the NSF CISE Advisory Committee and the Board of Directors of the Computing Research Association. She has conducted and published research in a number of subareas of Artificial Intelligence, including natural-language processing, automated plan generation and execution, and adaptive interfaces, and has been a pioneer in the application of AI methods to the design of assistive technology for people with cognitive impairment. Dr. Pollack is the recipient of a number of professional awards, including the Computers and Thought Award, the University of Pittsburgh Distinguished Research Award, and the Sarah Goddard Power Award.

Name: Prasad Tetali
Affiliation: School of Mathematics and College of Computing, Georgia Tech
Date: Wed, Dec 3, 2008
Time: 4:00 - 5:30pm
Host: Pete Winkler
Title: Correlation Decay and Deterministic Approximation Algorithms
Abstract: The notion of a correlation decay, originating in statistical physics, has recently played an important role in yielding fully polynomial time deterministic approximation algorithms for various counting problems. I will try to illustrate this technique (in a self-contained way) with two examples: counting matchings in bounded degree graphs, and counting independent sets in certain claw-free graphs.
Bio:

Name: Jim Haxby
Affiliation: Psychology and Brain Sciences, Dartmouth College
Date: Wed, Jan 14th, 2009
Host: Tanzeem Choudhury
Title: Characterizing local neural representation as a multidimensional similarity space
Abstract: Whereas conventional univariate analysis of functional brain imaging data characterized the function of a region in terms of the conditions that activate that region, multivariate pattern (MVP) analysis characterizes local function in terms of the conditions that evoke distinct patterns of activity. Moreover, the dissimilarities of the patterns of activity for different conditions can be quantified. Thus, local neural representation can be analyzed in terms of a high-dimensional similarity structure rather than as a (short) list of functions. Functional differences among brain regions can similarly by analyzed as differences in the neural representational space rather than as different functional labels. For example, different categories of visual stimuli – faces and objects – activate and evoke distinct patterns of activity in medial occipital, inferior lateral occipital (LO), and ventral temporal (VT) cortex, including when analysis is restricted to subregions that respond maximally to faces (FFA) and places (PPA). The similarity structure of the responses to categories, however, differs significantly among these brain regions. Whereas LO demonstrates larger distinctions than does VT within animate (faces of different species) and inanimate domains (houses, chairs, and shoes), VT demonstrates larger distinctions between the animate and inanimate domains. Medial occipital cortex, on the other hand, demonstrates a similarity structure that is not dominated by the animate-inanimate distinction at all. MVP analysis, therefore, reveals how local neural representation projects information into different subspaces that emphasize different distinctions among conditions. These methods provide a powerful tool for investigating how information is processed and re-represented in hierarchical and distributed neural systems.
Bio: Jim Haxby is the Director of the Center for Cognitive Neuroscience at Dartmouth. Prior to Dartmouth, he worked at the National Institutes of Health for twenty years followed by five years at Princeton University. He is a cognitive neuroscientist with an interest in face and object perception. Recently, his work has concentrated on the application of pattern classification methods from machine learning to the analysis of functional brain imaging data.

Name: Michael Casey
Affiliation: Electro-Acoustic Music, Dartmouth College
Date: Wed, Jan 21st, 2009
Location: Silsby 028 - note special location
Host: Tanzeem Choudhury
Title: The Problem with Music: Modeling Distance Distributions of Large Music Collections
Abstract: Recently, a number of piano recordings by different artists were found in a classical music catalog that exhibited a striking resemblance to each other. Could this resemblance be purely coincidental? We set about building a system that could answer this question and others in large recorded collections of music. The AudioDB system listens to polyphonic music recordings and encodes important perceptual information about them at fine time scales. The information extracted corresponds to traditional music-theoretic concepts such as, harmony, timbre, pitch, texture and rhythm yielding a high-dimensional representation; consisting of 300-1200 dimensions.
Our music databases have 10⁴-10⁷ recordings, each with thousands of vectors, so brute-force methods for similarity computation are not practical. Instead, we use locality-sensitive hashing (LSH) which searches in high dimensions with sub-linear time complexity. We propose a method for automatically estimating the radius threshold for efficient and accurate LSH retrieval. Our method employs statistical sampling of the background distance distribution and solving for the minimum distance distribution using order statistics.
Using these methods, we are able to quantify the "purely coincidental" resemblance in the piano recordings mentioned above, demonstrating that their similarity is not, in fact, coincidental. The newer recordings are altered copies of the older ones. Detecting fraudulent recordings with human hearing is difficult; even the music critics were fooled into highly commending these newer recordings. We conclude that scalable audio search systems, such as AudioDB, are required to address the emerging multimedia needs of the Internet, commercial music services and large multimedia archives.
Bio: Professor Michael Casey is director of the graduate program in Digital Musics at Dartmouth College. He received his Ph.D. from the MIT Media Laboratory in 1998 and has since authored numerous articles in the fields of music information retrieval, statistical audio analysis/synthesis, and audio-visual signal processing. Prior to coming to Dartmouth he was Professor of Computer Science at the University of London, where he lead a music computing research group, and Research Scientist for Mitsubishi Electric Research Laboratories (MERL) working on non-speech audio analysis and classification. Michael currently leads the development of AudioDB, an open source initiative, in collaboration with Dartmouth Digital Musics, OMRAS2 at Goldsmiths, University of London, and Yahoo! Research Inc. Support for this research was provided by the UK Engineering and Physical Sciences Research Council (EPSRC) and a Yahoo! Research Alliance award.

Name: Aurel Lazar
Affiliation: Electrical Engineering, Columbia University
Date: Wed, Jan 28th, 2009
Location: Silsby 028 - note special location
Host: Andrew Campbell
Title: Invariant Representations of Visual Streams in the Spike Domain
Abstract: We investigate an architecture for invariant representations of visual stimuli such as natural and synthetic video streams (movies, animation) in the spike domain. The stimuli are encoded with a population of spiking neurons, processed in the spike domain and finally decoded. The population of spiking neurons includes level crossing as well as integrate-and-fire neuron models with feedback. A number of spike domain processing algorithms are demonstrated including faithful stimulus recovery, as well as simple operations on the original visual stimulus such as translations, rotations and zooming. All these operations are executed in the spike domain. Finally, the processed spike trains are decoded for the faithful recovery of the stimulus and its transformations.
We show that the class of linear operations described above can easily be realized with the same basic stimulus decoding algorithm. What changes in the architecture, however, is the switching matrix (i.e., the input/output ``wiring'') of the spike domain switching building block. For example, for a particular setting of the switching matrix, the original stimulus is faithfully recovered. For other settings, translations, rotations and dilations (or combinations of these operations) of the original video stream are obtained.
Bio: Aurel A. Lazar is a Professor of Electrical Engineering at Columbia University. In the mid 80s and 90s, he pioneered investigations into networking games and programmable networks. In addition, he conducted research in broadband networking with quality of service constraints; and in architectures, network management and control of telecommunications networks. His current research interests are at the intersection of Computational Neuroscience, Information/ Communications Theory and Systems Biology. In silico, his focus is on Time Encoding and Information Representation in Sensory Systems, and, Spike Processing and Neural Computation in the Cortex. In vivo, his focus is on the olfactory system of the Drosophila.

Name: Deepak Ganesan
Affiliation: Computer Science, UMass Amherst
Date: Wed, Feb 4th, 2009
Location: Silsby 028
Host: Andrew Campbell
Title: Towards a Storage-centric Sensor Network Architecture
Abstract: A significant amount of sensor network research has addressed the problem of energy-efficiency, primarily by exploiting the fact that local computation is considerably less expensive than wireless communication. However, NAND flash memories allow storage to be more than two orders of magnitude cheaper than communication, significantly altering existing tradeoffs. In this talk, I will present our research on the design of a "rich sensor data management stack" that aims to make the role of storage central in sensor systems. I will focus on two recent research efforts in this talk. First, I will describe the design of novel flash-optimized index structures for archival storage and querying of sensor data. Second, I will describe an embedded image search engine to enable local storage of images on camera sensors (or mobile phones), and fast and efficient distributed search across such a camera network.
Bio: Deepak Ganesan is an Assistant Professor at the University of Massachusetts Amherst. He received his B.Tech in Computer Science from the Indian Institute of Technology, Madras, India, in 1998 and his Ph.D. from the University of California, Los Angeles, 2004. His research interests are in sensor networks, wireless networks, and flash-based storage systems. He has served on the program committees of a number of top conferences including NSDI and SenSys, and is an editor for the ACM Transactions on Sensor Networks. He is a recipient of the NSF CAREER Award and the IBM Faculty Partnership Award, and was selected as a UMass Chancellor's Junior Faculty Fellow.

Name: Dieter Fox
Affiliation: Computer Science, University of Washington
Date: Wed, Feb 11th, 2009
Location: Filene Auditorium, Moore Hall
Host: Tanzeem Choudhury
Title: Toward High-level Reasoning for Autonomous Systems
Abstract: Over the last decade, the robotics community has developed highly efficient and robust solutions to state estimation problems such as robot localization, people tracking, and map building. With the availability of various techniques for spatially consistent sensor integration, an important next goal is to enable robots to reason about the many objects located in our everyday environments and to reason about spatial concepts such as rooms, hallways, streets, and intersections. An additional requirement for successful operation in populated environments is the ability to recognize the intent of humans and to adapt to their behavior patterns.
In this talk I will present an overview of some recent work using graphical models and machine learning techniques to extract high level information from raw sensor data. Examples include place and object recognition from vision and laser data, and human activity recognition from wearable sensor data. I will conclude with an outlook for research directions in autonomous systems.
Bio: Dieter Fox is Associate Professor and Director of the Robotics and State Estimation Lab in the Computer Science & Engineering Department at the University of Washington, Seattle. He obtained his Ph.D. from the University of Bonn, Germany. Before joining UW, he spent two years as a postdoctoral researcher at the CMU Robot Learning Lab. Dieter's research focuses on probabilistic state estimation with applications in robotics and activity recognition. He has published over 100 technical papers and is co-author of the text book "Probabilistic Robotics". Dieter is an Associate Editor of JAIR and was program co-chair of AAAI-08. He has received several awards for his research, including an NSF CAREER award and best paper awards at robotics and Artificial Intelligence conferences.

Name: Szymon Rusinkiewicz
Affiliation: Computer Science, Princeton University
Date: Wed, Feb 18th, 2009
Location: Silsby 028
Host: Fabio Pellacini
Title: A System for High-Volume Acquisition and Matching of Fresco Fragments: Reassembling Theran Wall Paintings
Abstract: The archaeological site of Akrotiri on the volcanic island of Thera (modern-day Santorini, Greece) has yielded an unparalleled trove of artifacts and information from the prehistoric Aegean. The ancient civilization was destroyed by a volcanic eruption, which buried the remains of a flourishing Late Bronze Age (c. 1630 B.C.) settlement in ash. Among the most significant finds are numerous wall paintings, ranging from naturalistic and narrative scenes to abstract motifs. However, these paintings are recovered as thousands of plaster fragments, and reassembling them consumes a substantial portion of the effort expended at Akrotiri.
I will describe a system that uses 3-D and 2-D digitization hardware, together with computer-based matching techniques, to assist archaeologists and conservators in documenting and reassembling the wall paintings. Although mature technologies exist for acquiring images, geometry, and surface normals of small objects, they remain cumbersome and time-consuming for non-experts to employ on a large scale. Our system addresses the scalability, usability, and quality challenges of large-scale 3-D and 2-D digitization, by incorporating new algorithms to automatically align 3-D scans, register 2-D scans to 3-D geometry, and compute surface normals from 2-D scans. A novel 3-D matching algorithm efficiently searches for matching fragments using the scanned geometric models.
Bio: Szymon Rusinkiewicz is an associate professor of Computer Science at Princeton University. His work focuses on acquisition and analysis of the 3D shape and appearance of real-world objects, including the design of capture devices, data structures for efficient representation, and applications (most notably to cultural heritage objects and human skin). He also investigates algorithms for processing complex datasets of shape and reflectance, including registration, matching, completion, symmetry analysis, and sampling. His research interests also include illustrative depiction through line-drawings and non-photorealistic shading models.

Name: Tom Mitchell
Affiliation: Machine Learning Department School of Computer Science, Carnegie Mellon University
Date: ~~Thursday Feb 19th, 2009~~ RESCHEDULED DUE TO WEATHER
Location: Filene Auditorium, Moore Hall
Host: Jointly sponsored by the Center for Cognitive Neuroscience and Computer Science
Title: Brains, Meaning and Corpus Statistics
Abstract: How does the human brain represent meanings of words and pictures in terms of the underlying neural activity? This talk will present our research using machine learning methods together with fMRI brain imaging to study this question. One line of our research has involved training classifiers that identify which word a person is thinking about, based on the image of their fMRI brain activity. A more recent line involves developing a generative computational model that predicts the neural activity associated with arbitrary English words, including words for which we do not yet have brain image data. This computational model is trained using a combination of fMRI data associated with several dozen concrete nouns, together with statistics gathered from a trillion-word text corpus. Once trained, the model predicts fMRI activation for any other concrete noun appearing in the tera-word text corpus, with highly significant accuracies over the 100 nouns for which we currently have fMRI data.
Bio: Tom Mitchell is the E. Fredkin Professor and Department Head Machine Learning Department Carnegie Mellon University.

Name: Laura Toma
Affiliation: Department of Computer Science, Bowdoin College
Date: Wed, Feb 25th, 2009
Location: Silsby 028
Host: Afra Zomorodian
Title: I/O-Efficient Indexes for Fat Triangulations and Low-Density Planar Subdivisions
Abstract: The traditional approach to algorithms design considers each atomic operation to take roughly the same amount of time. Unfortunately this assumption is not valid when the algorithm operates on data stored on disk: reading data from or writing data to disk can be a factor 1,000,000 or more slower than an operation on data that is already present in main memory. This has led to the study of so-called I/O-
efficient algorithms, which specifically optimize the number of blocks transfered between main memory and disk.

In this talk I will describe two I/O-efficient index structures for storing planar subdivisions. The structures are based on quadtrees and exploit features of realistic terrains: they are provably efficient when the input has low-density (any disk D is intersected by at most a constant number edges whose length is at least the diameter of D) or is a fat triangulation (every angle is bounded from below by a fixed constant), respectively.

In particular, we show that a low-density subdivision with n edges can be preprocessed i in $O(\sort(n))$ IOs, allowing to compute the intersections between the edges of two such preprocessed subdivisions in $O(scan(n))$ IOs, where n is the total number of edges in the two subdivisions; and allowing to answer a single point location query in $O(\log_B n)$ IOs.

The data structures improve on the previous best known bounds for overlaying subdivisions, both in the number of IOs and storage usage. Moreover, they are significantly simpler and they are cache- oblivious. preliminary experimental results have shown that they are fast and scalable in practice on real-world data.
Bio: Laura Toma is Assistant Professor at Bowdoin College since 2003. Laura is originally from Romania and got her PhD degree from Duke University in 2003. Her research is in the area of I/O-efficient algorithms and algorithm engineering, and in particular terrain processing and applications to GIS.

Name: Frank Bentley
Affiliation:Motorola Applied Research and Technology Center
Date: March 4th, 2009
Location: Silsby 028
Host: Andrew Campbell
Title: Ambient Mobile Micro-coordination: Putting Presence and People in the Loop
Abstract: Micro-coordination has traditionally been a tedious task. Deciding where to meet someone and when, and then actually getting there is a process that often involves multiple emails and phone calls and sometimes violent hand gestures to indicate that “I’m over here!” In the Motorola Social Media Research Lab, we’ve implemented and studied several mobile presence systems to help people coordinate with others in a much more passive manner. We’ve found that people can often infer quite a lot from a simple bit of presence information and can use that information in planning communication and in-person interactions. I’ll describe our systems as well as our field research and provide implications for the design of mobile social applications.
Bio: Frank Bentley is a Principal Staff Research Scientist in the Experiences Research Lab of the Motorola Applied Research and Technology Center in Chicago. He studies how people interact with new communications technology and is involved in all aspects of research projects from early ethonographic-style research through to design, prototyping, field trials, and commercialization. His research interests lie in the areas of ambient interfaces, mobile computing, and social media. Frank also teaches a multi-disciplinary class on “Communicating with Mobile Technology” in the Spring at MIT.

Name: Jim Kurose
Affiliation: Department of Computer Science, UMass Amherst
Date: April 1st, 2009
Host: Andrew Campbell
Title: Collaborative Adaptive Sensing of the Atmosphere: Challenges in End-to-End Sensor Networking
Abstract: The CASA project is an NSF Engineering Research Center investigating the design and implementation of a dense network of low-power meteorological radars whose goal is to collaboratively and adaptively sense the lowest few kilometers of the earth's atmosphere. In the first part of this talk we overview the CASA project, describe its computing and networking challenges, and overview the software/network architecture and implementation of the CASA testbeds. We also discuss the operation of CASA’s testbed during the spring tornado season in Oklahoma. In the second part of this talk, we focus on networking-related research issues and discuss our experiences in using user-specified preferences to drive the optimization of the network's radar scanning behavior. Throughout the talk, we’ll discussion of a number of interesting on-going and open research issues.
Bio: Jim Kurose received a B.A. degree in physics from Wesleyan University and his Ph.D. degree in computer science from Columbia University. He is currently Interim Dean of the College of Natural Science and Mathematics and Distinguished University Professor (and past chairman) in the Department of Computer Science at the University of Massachusetts. He is also co-director of the Networking Research Laboratory and Associate Director of the NSF Engineering Research Center for Collaborative Adaptive Sensing of the Atmosphere (CASA). Professor Kurose has been a Visiting Scientist atIBM Research, INRIA, Institut EURECOM , the University of Paris, LIP6, and Thomson Research Labs. His research interests include network protocols and architecture, network measurement, sensor networks, multimedia communication, and modeling and performance evaluation. Dr. Kurose has served as Editor-in-Chief of the IEEE Transactions on Communications and was the founding Editor-in-Chief of the IEEE/ACM Transactions on Networking. He has been active in the program committees for IEEE Infocom, ACM SIGCOMM, and ACM SIGMETRICS conferences for a number of years, and has served as Technical Program Co-Chair for these conferences. He has won several conference best paper awards and received the ACM Sigcomm Test of Time Award.

Name: David Fleet
Affiliation: Department of Computer Science, University of Toronto
Date: April 8th, 2009
Host: Lorenzo Torresani
Title: Model-Based Human Pose Tracking
Abstract: Future computer vision systems will recognize people and their activities, enabling myriad new applications, such as smart mobile devices, advanced surveillance systems, and new perceptual man-machine interfaces. The detection, tracking and pose estimation of people from digital images are therefore key problems in Computer Vision. Pose estimation is especially challenging because image measurements are often insufficient to fully constrain 3D pose, and as a consequence, current formulations rely heavily on prior models of human pose and model.

This talk presents two new techniques for modeling human pose and motion. The first is a probabilistic latent variable model called the Gaussian Process Dynamical Model. It is a form of probabilistic nonlinear dimensionality reduction for time-series data. The second class of models stems from Newtonian physics and biomechanical principles. The formulation of these models and applications to visual tracking will be described. Where time permits I will also mention several other emerging research directions that appear promising.

Bio: David Fleet is professor of computer science at the University of Toronto. He received the PhD in Computer Science from the University of Toronto in 1991. From 1991 to 2000 he was on faculty at Queen's University, Canada, in the Department of Computing and Information Science, with cross-appointments in Psychology and Electrical Engineering. In 1999 he joined the Palo Alto Research Center (PARC) where he managed the Digital Video Analysis Group and the Perceptual Document Analysis Group. He returned to the University of Toronto in October 2003.

In 1996 Dr. Fleet was awarded an Alfred P. Sloan Research Fellowship for his research on biological vision. His 1999 paper with Michael Black on probabilistic detection and tracking of motion boundaries received Honorable Mention for the Marr Prize at the IEEE International Conference on Computer Vision. His 2001 paper with Allan Jepson and Thomas El-Maraghi on robust appearance models for visual tracking was awarded runner-up for the best paper at the IEEE Conference on Computer Vision and Pattern Recognition. In 2003, his paper with Eric Saund, James Mahoney and Dan Larner won the best paper award at ACM UIST '03. He was Associate Editor of IEEE Transactions on
Pattern Analysis and Machine Intelligence (2000-2004), Program Co-Chair for the IEEE Conference on Computer Vision and Pattern Recognition in 2003, and Associate Editor-In-Chief for IEEE Transactions on Pattern Analysis and Machine Intelligence (2005-2008). He is Fellow of the Canadian Institute of Advanced Research.

His research interests include computer vision, image processing, visual perception, and visual neuroscience. He has published research articles and one book on various topics including the estimation of optical flow and stereoscopic disparity, probabilistic methods in motion analysis, 3D people tracking, modeling appearance in image sequences, non-Fourier motion and stereo perception, and the neural basis of stereo vision.

Name: Bruce Shepherd CANCELLED
Affiliation: Department of Mathematics and Statistics,McGill University
Date: Wed, April 15, 2009
Time: 4:00 - 5:30pm
Host: Lisa Fleischer
Title: Coming Soon
Abstract:
Bio:

Name: Bill Freeman
Affiliation: CSAIL , MIT
Date: POSTPONED TILL FALL 2009
Time: 4:00 - 5:30pm
Host: Lorenzo Torresani
Title: Coming Soon
Abstract:
Bio:

Name: Tom Mitchell
Affiliation: Machine Learning Department School of Computer Science, Carnegie Mellon University
Date: Thursday, April 23, 2009 Please note that this talk will be on Thursday instead of our regular Wednesday seminars
Time: 4:00 - 5:30pm
Location: Rockefeller 001
Host: Jointly sponsored by the Center for Cognitive Neuroscience and Computer Science
Title: Brains, Meaning and Corpus Statistics
Abstract: How does the human brain represent meanings of words and pictures in terms of the underlying neural activity? This talk will present our research using machine learning methods together with fMRI brain imaging to study this question. One line of our research has involved training classifiers that identify which word a person is thinking about, based on the image of their fMRI brain activity. A more recent line involves developing a generative computational model that predicts the neural activity associated with arbitrary English words, including words for which we do not yet have brain image data. This computational model is trained using a combination of fMRI data associated with several dozen concrete nouns, together with statistics gathered from a trillion-word text corpus. Once trained, the model predicts fMRI activation for any other concrete noun appearing in the tera-word text corpus, with highly significant accuracies over the 100 nouns for which we currently have fMRI data.

Bio: Tom Mitchell is the E. Fredkin Professor and Department Head Machine Learning Department Carnegie Mellon University.

Name: Luis Von Ahn
Affiliation: Computer Science , CMU
Date: Wed, April 29, 2009
Time: 4:00 - 5:30pm
Host: Jointly sponsored by Digital Humanities and Computer Science
Title: Human Computation
Abstract: This talk is about harnessing human brainpower to solve problems that computers cannot. Although computers have advanced dramatically over the last 50 years, they still do not possess the basic conceptual intelligence or perceptual capabilities that most humans take for granted. By leveraging human abilities in a novel way, I want to solve large-scale computational problems and collect data to teach computers basic human talents. To this end, I treat human brains as processors in a distributed system, each performing a small part of a massive computation. Unlike computer processors, however, humans require an incentive to join a collective computation. Among other things, I show how to use online games as a means to encourage participation in the process.
Bio: Professor Luis von Ahn works in the Computer Science Department at Carnegie Mellon University. He is the recipient of a MacArthur Fellowship, a Sloan Fellowship, and a Microsoft New Faculty Fellowship. He has been named one of the 50 Best Minds in Science by Discover Magazine, one of the "Brilliant 10" of 2006 by Popular Science Magazine, one of the 50 most influential people in technology by Silicon.com, and one of the Top Innovators in the Arts and Sciences by Smithsonian Magazine. His research interests include encouraging people to do work for free, as well as catching and thwarting cheaters in online environments.

Name: George Cybenko
Affiliation: Thayer School of Engineering, Dartmouth College
Date: Wed, May 6, 2009
Time: 4:00 - 5:30pm
Host: Andrew Campbell
Title: Learning Behaviors: The next big (computational) thing?
Abstract: Learning, analyzing and detecting behaviors of computers, networks, people and human organizations is a growing multi-disciplinary area. This talk will survey some recent algorithmic and applications results, with a focus on ongoing projects at the Thayer School.
Bio: George Cybenko is the Dorothy and Walter Gramm Professor of Engineering at the Thayer School of Engineering at Dartmouth. Prior to joining the Dartmouth faculty in 1992, he was Professor of Electrical and Computer Engineering and Computer Science at the University of Illinois at Urbana-Champaign. Cybenko's current research interests are behavioral modeling and analysis in various technical areas. Cybenko was the founding Editor-in-Chief of IEEE/AIP Computing in Science and Engineering and IEEE Security & Privacy. He serves on the Defense Science Board, the IEEE Computer Society's Board of Governors and the Computing Research Association's Board of Directors. Cybenko received a B.Sc. degree from the University of Toronto and a Ph.D. degree Princeton, both in Mathematics, and is a Fellow of the IEEE.

Name: Katherine Isbister
Affiliation: NYU-Poly and ITU Copenhagen's Center for Computer Games Research
Date: Wed, May 13, 2009
Time: 4:00 - 5:30pm
Host: Mary Flanagan (Jointly sponsored by Digital Humanities and Computer Science)
Title: Touchy Feely Games: Broadening Designers' Social and Emotional Palette
Abstract: In the 1980s the fledgling Electronic Arts company asked 'Can a Computer Make You Cry?' in a famous advertising campaign, and set a challenge for the field that many would say has yet to be fully realized. The popular press often emphasizes violent or 'primitive' emotions in its portrayal of games, and holds up the solitary antisocial gamer as a warning to us all. Yet the Entertainment Software Association reports that over half of gaming is done socially, and game designers and developers have recently renewed enthusiasm and efforts devoted to making games more emotionally rich and appealing. I believe that a wide range of social and emotional qualities can be designed into games and other interactive experiences, and will present research projects that show my efforts to understand what drives emotion and social connection, and that apply what I've learned to create designed experiences that push the boundaries of the medium toward a broader social and emotional palette.
Bio: Katherine Isbister is an Associate Professor of Digital Media and Computer Science and Engineering at NYU-Poly, and also maintains an affiliation at the ITU Copenhagen Center for Computer Games Research. She received her Ph.D. from Stanford University, with a focus on applying human social behavior to the design of digital characters. Since then, she has worked in both industry and research venues to create and evaluate interfaces that enhance the player (or user) experience using social and emotional qualities.

Dr. Isbister has written two books: Better Game Characters by Design: A Psychological Approach, and Game Usability: Advice from the Experts for Advancing the Player Experience. Better Game Characters was nominated for a Game Developer Magazine Frontline Award in 2006.

Current research interests include emotion and gesture in games, supple interactions, design of game characters, and game usability. Dr. Isbister presents and publishes her work in Human Computer Interaction (HCI) and Game Studies venues. She has received funding from both governmental and private sources, including the U.S. National Science Foundation and companies such as Electronic Arts. She serves on the advisory board of the International Game Developers Association Games Education Special Interest Group, and on the Editorial Board of the International Journal of Human Computer Studies. In 1999 Katherine Isbister was selected as one of MIT Technology Review's TR100 Young Innovators most likey to shape the future of technology.

Name: Rosalind Picard
Affiliation: Media Laboratory, MIT
Date: Wed, May 20, 2009
Time: 4:00 - 5:30pm
Host: Tanzeem Choudhury (Jointly sponsored by Digital Humanities and Computer Science)
Title: Emotional Intelligence Technology and Autism
Abstract: Skills of emotional intelligence include the ability to recognize and respond appropriately to another person's emotion, and the ability to know when (not) to display emotion. This talk will demonstrate advances at MIT aimed at giving several of these skills to technology including mobile devices, robots, agents, wearable & traditional computers. I will present a live demonstration of current technology, developed w/el Kaliouby, to recognize complex cognitive-affective states in real time from a person's head and facial movements. This technology computes probabilities that a person looks like he or she is concentrating, interested, agreeing, disagreeing, confused, or thinking. These states signal important information such as when is a good time to interrupt, or when might be appropriate to apologize for interrupting. A wearable version of this system is being developed for helping people with autism who often face challenges reading social-emotional cues. I will describe several other new affective technologies that facilitate emotion measurement and communication, and highlight social, ethical, and philosophical issues surrounding their use.
Bio: Professor Rosalind W. Picard is the founder and director of the Affective Computing Research Group at the Massachusetts Institute of Technology (MIT) Media Laboratory, co-director of the Things That Think Consortium, the largest industrial sponsorship organization at the lab, and leader of the new and growing Autism Communication Technology Initiative at MIT. In April 2009 she co-founded Affectiva with Dr. Rana el Kaliouby, to commercialize technologies for emotion measurement and communication. Picard holds a Bachelors in Electrical Engineering with highest honors from the Georgia Institute of Technology, and Masters and Doctorate degrees, both in Electrical Engineering and Computer Science, from the Massachusetts Institute of Technology (MIT). Prior to completing her doctorate at MIT, she was a Member of the Technical Staff at AT&T Bell Laboratories where she designed VLSI chips for digital signal processing and developed new methods of image compression and analysis. In 1991 she joined the MIT Media Lab faculty, where she became internationally known for constructing powerful mathematical models for content-based retrieval of images, for creating new tools such as the Photobook system, and for pioneering methods of automated search and annotation in digital video. The year before she was up for tenure, she published the award-winning book Affective Computing, which was instrumental in starting a new field by that name. Picard has been awarded dozens of distinguished and named lectureships internationally and in 2005 was honored as a Fellow of the IEEE for contributions to image and video analysis and affective computing.

Name: Kimo Johnson
Affiliation:MIT
Date: Wed, May 27, 2009
Time: 4:00 - 5:30pm
Host: Hany Farid
Title: Coming Soon
Abstract:
Bio: