Fall 2012
Student Seminar
September 17, 2012
Automatic human activity analysis from videos is a very important area of research in computer vision with numerous applications such as surveillance, security, human machine interfaces, sports training, elderly care and monitoring, building smart home and public environments as well as mining web videos for activity based content. From a research point of view, of particular interest are the development of a) rich representations for human motion across several domains, 2) algorithms for tasks such as action recognition and tracking, and c) computationally efficient strategies for performing human activity analysis in very large data sets. In this talk we will address these challenges and propose features and methods for human activity analysis that are very general and can be applied across several domains. The common thread underlying all the methods is the need to explicitly model the temporal dynamics of human motion.
We will first propose optical-flow based and medial-axis skeletal time series features to represent human motion in a scene. We will then model the temporal evolution of these feature time-series using dynamical systems with stochastic inputs and develop methods for comparing these dynamical systems for the purpose of human activity recognition. We will then address the issue of human activity tracking by proposing action-specific dynamic templates. The tracking problem will be posed as a joint optimization problem over the location of the person in the scene as well as the internal state of the dynamical system driving the particular activity. Finally, we will propose very fast approximate-nearest neighbor based methods on the space of dynamical systems for analyzing human motion and show that we can perform human activity recognition very efficiently albeit at a cost of slightly decreased accuracy. Our experimental analysis will show that using dynamical-systems based generative models for human activity perform very well in the above-mentioned tasks.
Speaker Biography: Rizwan Chaudhry received the BSc. (Honors) degree in 2005 with a double major in Computer Science and Mathematics from the Lahore University of Management Sciences in Lahore, Pakistan. Since 2006, he has been pursuing a Ph.D. in the Department of Computer Science at the Johns Hopkins University, where he has been associated with the Vision, Dynamics and Learning Lab under the guidance of Dr. Rene Vidal. His research interests are in the general areas of computer vision and machine learning, and more specifically, modeling dynamic visual phenomena, human activity recognition and tracking.
September 20, 2012
Molecular dynamics computer calculations with modern hardware applied to biological molecules can now reach trajectory lengths up to microseconds in length. Yet, to couple these simulations to grand-challenge problems in medicine and biology: genomics, metabolomics, signaling events, and drug design, still more sampling of the conformational space available to biomolecules is needed. This creates challenges for computer algorithms and data structures to go beyond the current approach and find new ways to sample and to organize the information.
This talk will describe two algorithms and recent work in data-structures that may address the sampling problems. The dynamic importance sampling method grows out of Monte Carlo importance sampling to dynamically enhance sampling of intermediate conformations between stable states. The effective transfer entropy method may provide reduced dimensionality projections to further enhance sampling. These directions are coupled with suggestions for control and hierarchy in data-structures to aid the more efficient and information-rich collection of molecular trajectories.
Speaker Biography: Tom Woolf has a physics background, with a BS in Physics from Stanford, MS from the University of Chicago, and a PhD from Yale University in Biophysics/Neuroscience. His post-doctoral work was on molecular dynamics simulations of the gramicidin channel and he’s been pursuing molecular dynamics simulations since arriving at Hopkins in 1994. He is currently a Professor at the School of Medicine in the Department of Physiology. His research has pursued studies aimed at structure:function questions for membrane proteins, at relative free energy calculations for ligand binding, and at the kinetics of conformational change. This later research direction has led to his interest in how to most efficiently explore and understand conformational space.
September 25, 2012
MRI (Magnetic Resonance Image) has been recognized as an established image guidance tool for a minimally invasive therapy in recent years. We have developed a novel MR -image guided surgical system for endoscopic surgery in the closed bore MRI. Using a new real-time MR image guided surgical system, endoscopic surgical procedures can be performed with endoscopic images and real-time MR image combination. In this talk, I would like to present our new surgical system of MR image-guidance and a concept of clinical applications.
Speaker Biography: Shigeyuki Naka is lecturer of Department of Surgery, Shiga University of Medical Science, Japan. His research interests are Minimally Invasive Surgery, MR Image Guided Therapy and Robotic Surgery.
September 25, 2012
For the last decade, a various kinds of energy (ultrasonic and radiofrequency) surgical devices have been developed, and they have supported the progression of less-invasive surgery. However, majority surgeons are not satisfied with these devices, because that the capability of vessel sealing is not adequate to perform various surgical procedures with one device. We have originally invented new energy devices using microwave energy for sealing and cutting of vessels as well as solid organs, and brought all surgical procedures into completion with only one device. In future, microwave surgical devices may be applied to all kinds of less-invasive and robotic surgery. In this talk, I would like to present a mechanism of microwave heating for live tissue and applications to less-invasive surgery.
Speaker Biography: Tohru Tani is Professor and Chairman of department of surgery, Adviser of President and Director of Biomedical Innovation Center, Shiga University of Medical Science. His research interests are Cancer therapy for digestive organs, Artificial Organs, MR Image Guided Therapy and Robotic Surgery.
October 2, 2012
Large amounts of multivariate time series data are now being collected for tracking phenomena that evolve over time. Approaches that can incorporate expert biases to discover informative representations of such data are valuable in enabling exploratory analysis and feature construction.
In this talk, we will discuss priors to cluster time series that leverage two different sets of assumptions frequently present in real world data. First, we want to identify repeating “shapes”. For example, when we walk vs. kick, our joint angles produce different repeating shapes over time. How do we discover such repeating shapes in the absence of any labeled data? The second assumption we will tackle is when the generated data does not come from a fixed set of one or two classes as is the case in binary prediction. For example, in clinical data, no two patients are alike. Assuming that generated data is sampled from an m-class distribution can inappropriately bias your learned model.
I will show results from multiple domains but will focus on a novel application of modeling physiologic data from monitoring infants in the ICU. Insights gained from our exploratory analysis led to a novel risk prediction score that combines patterns from continuous physiological signals to predict which infants are at risk for developing major complications. This work was published on the cover of Science Translational Medicine (Science’s new journal aimed at translational medicine work), and was covered by numerous national and international press sources.
Speaker Biography: Suchi Saria recently joined Johns Hopkins as an Assistant Professor in the departments of Computer Science and Health Policy & Management. She received her PhD in Computer Science from Stanford University working with Prof. Daphne Koller. She has won various awards including, a Best Student Paper awards, the Rambus Fellowship, the Microsoft full scholarship and the National Science Foundation Computing Innovation Fellowship. Prof. Saria’s interests are in graphical models, machine learning with applications in modeling complex temporal systems. In particular, she wants to help solve the trillion dollar question of how to fix our health care system and show how these approaches can help us improve the delivery of healthcare.
October 18, 2012
Deep learning and unsupervised feature learning offer the potential to transform many domains such as vision, speech, and NLP. However, these methods have been fundamentally limited by our computational abilities, and typically applied to small-sized problems. In this talk, I describe the key ideas that enabled scaling deep learning algorithms to train a very large model on a cluster of 16,000 CPU cores (2000 machines). This network has 1.15 billion parameters, which is more than 100x larger than the next largest network reported in the literature.Such network, when applied at the huge scale, is able to learn abstract concepts in a much more general manner than previously demonstrated. Specifically, we find that by training on 10 million unlabeled images, the network produces features that are very selective for high-level concepts such as human faces and cats. Using these features, we also obtain significant leaps in recognition performance on several large-scalecomputer vision tasks.
Speaker Biography: Quoc Le is a PhD student at Stanford and software engineer at Google. At Stanford and Google, Quoc works on large scale brain simulation using unsupervised feature learning and deep learning. His recent work was widely distributed and discussed on various technology blogs and news sites. Quoc obtained his undergraduate degree at Australian National University, and was research visitors at National ICT Australia, Microsoft Research and Max Planck Institute of Biological Cybernetics. Quoc won the best paper award as ECML 2007.
Student Seminar
October 19, 2012
As life expectancy in the industrialized world increases, so does the number of elders with chronic health conditions such as diabetes and congestive heart failure that require complex self-management routines. The traditional model of episodic care in clinic and hospital-based settings is not optimal for improving chronic disease outcomes. To solve such issues, we are acquiring tools to conduct continuous monitoring of people’s everyday lives, called pervasive health devices. Likewise, we are getting Web/mobile applications that analyze and visualize the data from pervasive health devices for users are called as pervasive health applications.
However, existing pervasive health devices and applications pose the following challenges to be effective in practice. First, they lack caregivers’ connection and feedback. Due to the absences, people tend to lose their interests in using devices quickly and they do not receive professional advice on their health. More importantly, the existing pervasive systems are designed closed and vertically-integrated, thereby impeding the development of sophisticated pervasive health applications for complex, multi-factorial disease and health conditions. Furthermore, the closed systems complicate unified management of pervasive health devices.
In this talk, I introduce a closed-loop system to solve the challenges in pervasive healthcare. The closed-loop consists of multiple software and hardware components through which device users receive in-depth remote cares and caregivers minimize the time and cost spent for healthcare. After I introduce the overall concept of closed-loop design and the potential impacts that it can make on caregivers and device users, I focus on HealthOS: a platform to develop pervasive health applications and also a core system to integrate all other components in our close-loop. Specifically, I describe how HealthOS simplifies application development and integrates heterogeneous hardware and software components, as well as its impact on pervasive healthcare.
Speaker Biography: Jong Hyun Lim is a Ph.D. candidate in the Department of Computer Science at the Johns Hopkins University and also a member of the Hopkins InterNetworking Research Group (HiNRG) led by his adviser, Dr. Andreas Terzis. His interest includes protocol design, system support, and target localization for healthcare systems.
Student Seminar
October 19, 2012
Over the past decade, many embedded sensing networks have been deployed to pervasively monitor the physical world at previously infeasible or impractical spatial and temporal resolutions. The result of this novel sensing ability is deeper understanding and higher degree of control of the physical world. The decreasing cost and size of the sensor nodes will enable even larger scales of embedded sensing systems and realize many more applications. However, this continued evolution poses numerous research challenges that have not been adequately addressed before. This dissertation proposes and evaluates solutions to some of those challenges, and aims to bring the vision of pervasive inter-connectivity among every physical objects in our lives a step closer to reality. The first part of this dissertation develops methods to mitigate the unreliable nature of low power radio communications, including a lightweight calibration mechanism for IEEE 802.15.4 radios, a framework for leveraging radio spatial diversity, and a distributed algorithm that is robust to packet losses. The second part of this dissertation designs an ultra low-power time synchronization module and develops an FM fingerprinting based indoor localization system, to provide time and location services for embedded sensing systems.
Speaker Biography: Yin Chen received the Bachelors in Engineering degree in Automation from Tsinghua University in 2007. In Fall 2007, he enrolled in the Ph.D. program in the Department of Computer Science at the Johns Hopkins University. At Johns Hopkins, Yin Chen was a member of the Hopkins interNetworking Research Group (HiNRG) led by Dr. Andreas Terzis. His research interests include application development, protocol design, information processing, and time synchronization in networked embedded sensing systems, as well as indoor localization, tracking, and human-centric sensing with mobile devices.
October 25, 2012
With the invention of DNA sequencing came an explosion of sequence data, and a lot of work for computer scientists. With the advent of the human genome project came another explosion of sequence data, and a lot more work for computer scientists. In just the last five years, we have seen another huge breakthrough: the invention of second-generation DNA sequencing. DNA sequencers now can generate hundreds of billions of nucleotides of data, enough to cover the human genome hundreds of times over, in about a week for a few thousand dollars. Consequently, sequencing has become a very common tool in the study of biology, genetics, and disease.
But with these developments comes a problem: growth in per-sequencer throughput is drastically outpacing growth in computer speed. As the gap widens over time, the crucial research bottlenecks are increasingly computational: computing, storage, labor, power. Life science simply cannot advance without help from computer science.
I will survey some current computational problems in genomics, and discuss a few solutions. My goal is to provide background, convey what kinds of computational problems are important in the field currently, and to highlight problems that cut across computational disciplines.
Speaker Biography: Ben Langmead recently joined Johns Hopkins as an Assistant Professor in the department of Computer Science. He received a Ph.D. in Computer Science from the University of Maryland working with Prof. Steven L Salzberg. Prof. Langmead uses approaches from computer science — algorithms, text indexing, and high performance computing, especially cloud computing — to create high-impact software tools for life science researchers. His goal is to make all types of high-throughput biological data, especially sequencing data, easy to analyze and interpret. Software tools written by Prof. Langmead, including Bowtie and Crossbow, are widely used in the genomics community. His work on Bowtie won the Genome BIology award for outstanding publication in the year 2009.
November 1, 2012
There has been remarkable progress in a field known as “derandomization” — leading to a situation where most experts now believe that any probabilistic algorithm can be replaced by a deterministic algorithm with comparable complexity. The tools and techniques of derandomization have also opened new connections between two fields that had previously seemed to have little connection to each other: 1. the field of circuit complexity (in which we try to find the most efficient circuitry to compute a given Boolean function), and 2. the field of algorithmic information theory (aka “Kolmogorov Complexity”), which provides a mathematical definition of “randomness”.
This lecture will introduce the listener to these two fields, and show how the study of derandomization has opened links that have enriched each field. In particular, the lecture will present theorems that suggest that two familiar complexity classes: * BPP: the class of problems that can be solved with negligible error in polynomial-time, and * NEXP: the class of problems solvable in nondeterministic exponential time can be characterized in terms of efficient reductions to the (undecidable) set of Kolmogorov-random strings.
Here are links to some papers related to this talk:
- http://ftp.cs.rutgers.edu/pub/allender/curiouser.pdf
- http://ftp.cs.rutgers.edu/pub/allender/ttrt.pdf
Speaker Biography: Eric Allender is widely recognized as a leading researcher in computational complexity theory. He has given numerous invited addresses at computer science symposia worldwide. He received a B.A. from the University of Iowa in 1979, majoring in Computer Science and Theater, and a Ph.D. from Georgia Tech in 1985. He has been at Rutgers University since then, serving as department chair from 2006 to 2009. He is a Fellow of the ACM, is Editor-In-Chief of ACM Transactions on Computation Theory, and serves on the editorial boards of Computational Complexity, and The Chicago Journal of Theoretical Computer Science. During the 2009/2010 academic year, he was a Fulbright Fellow in South Africa.
November 2, 2012
If you want to program a parallel computer, it obviously makes sense to start with a computational paradigm in which parallelism is the default (ie functional programming), rather than one in which computation is based on sequential flow of control (the imperative paradigm). And yet, and yet … functional programmers have been singing this tune since the 1980s, but do not yet rule the world. In this talk I’ll say why parallelism is too complex a beast to be slain at one blow, and how we are going to be driven, willy-nilly, towards a world in which side effects are much more tightly controlled than now – and this is a space where Haskell really shines. I’ll give a whirlwind tour of a whole range of ways of writing parallel program in a functional paradigm (implicit parallelism, transactional memory, data parallelism, DSLs for GPUs, distributed processes, etc, etc), illustrating with examples from the rapidly moving Haskell community, and identifying some of the challenges we need to tackle.
Speaker Biography: Simon Peyton-Jones, MA, MBCS, CEng, graduated from Trinity College Cambridge in 1980. After two years in industry, he spent seven years as a lecturer at University College London, and nine years as a professor at Glasgow University, before moving to Microsoft Research (Cambridge) in 1998. His main research interest is in functional programming languages, their implementation, and their application. He has led a succession of research projects focused around the design and implementation of production-quality functional-language systems for both uniprocessors and parallel machines. He was a key contributor to the design of the now-standard functional language Haskell, and is the lead designer of the widely-used Glasgow Haskell Compiler (GHC). He has written two textbooks about the implementation of functional languages. More generally, he is interested in language design, rich type systems, software component architectures, compiler technology, code generation, runtime systems, virtual machines, and garbage collection. He is particularly motivated by direct use of principled theory to practical language design and implementation — that’s one reason he loves functional programming so much.
Distinguished Lecturer
November 8, 2012
A general theory of trust in networks of humans and computers must be built on both a theory of behavioral trust and a theory of computational trust.1 This argument is motivated by increased participation of people in online social networking, crowdsourcing, human computation, and socio-economic protocols, e.g., protocols modeled by trust and gift-exchange games, norms-establishing contracts, and scams/deception. We illustrate a class of interactive social protocols that relies both on trustworthy properties of commodity systems2 (e.g., verifiable end-to-end trusted path) and participant trust, since on-line verification of protocol compliance is often impractical; e.g., it can lead to undecidable problems, co-NP complete test procedures, and user inconvenience. Trust is captured by participant preferences (i.e., risk and betrayal aversion) and beliefs in the trustworthiness of other protocol participants. Both preferences and beliefs can be enhanced whenever protocol non-compliance leads to punishment of untrustworthy participants; i.e., it seems natural that betrayal aversion can be decreased and belief in trustworthiness increased by properly defined punishment. Similarly, it seems natural that risk aversion can be decreased and trustworthiness increased by feasible recovery from participant non-compliance.
A general theory of trust which focuses on the establishment of new trust relations where none were possible before would help create new economic opportunities. New trust relations would increase the pool of services available to users, remove cooperation barriers, and enable the “network effect” where it really matters; i.e., at the application level. Hence, it seems important that security research should enable and promote trust-enhancement infrastructures in human and computer networks; e.g., trust networks. Finally, we argue that a general theory of trust should mirror human expectations and mental models of trust without relying on false metaphors and analogies with the physical world.
Speaker Biography: Virgil D. Gligor received his B.Sc., M.Sc., and Ph.D. degrees from the University of California at Berkeley. He taught at the University of Maryland between 1976 and 2007, and is currently a Professor of Electrical and Computer Engineering at Carnegie Mellon University and co-Director of CyLab. Over the past thirty-five years, his research interests ranged from access control mechanisms, penetration analysis, and denial-of-service protection to cryptographic protocols and applied cryptography. Gligor was an editorial board member of several IEEE and ACM journals, and the Editor in Chief of the IEEE Transactions on Dependable and Secure Computing. He received the 2006 National Information Systems Security Award jointly given by NIST and NSA in the US, and the 2011 Outstanding Innovation Award given by the ACM Special Interest Group on Security, Audit and Control.
Student Seminar
November 12, 2012
While robots routinely perform complex assembly tasks in highly structured factory environments, it is challenging to apply completely autonomous robotic systems in less structured manipulation tasks, such as surgery and machine assembly/repair, due to the limitations of sensor data interpretation, task modeling and machine intelligence. A practical, yet effective, approach to accomplish these tasks is through human-robot collaboration, in which the human operator and the robot form a partnership and complement each other in performing a complex task. We recognize that humans excel at determining task goals and recognizing constraints, if given sufficient feedback about the interaction between the tool (e.g., end-effector of the robot) and the environment. Robots are precise, unaffected by fatigue and able to work in environments not suitable for humans. Our research hypothesis is that human-robot collaboration can be improved via a system architecture that facilitates the bidirectional flow of information and assistance between the human and robot. In particular, the system provides information via visual and force (haptics) feedback, thereby enabling the operator to provide information and assistance to the system by defining the task model, in terms of task goals and virtual fixture constraints, through an interactive, or immersive, augmented reality interface. This then allows the robot system to actively assist the operator, to enhance the execution time, quality and precision of the tasks. We validate our hypothesis by implementing this system architecture for two different robot control paradigms, cooperative (i.e., hands-on) and telerobotic control, in two different application domains, image-guided robotic neurosurgery and telerobotic satellite servicing under significant time delay.
Speaker Biography: Tian Xia received the B.S.E degree in Electrical Engineering from the University of Washington in Seattle, and enrolled in the Computer Science Ph.D. program at the Johns Hopkins University. He received the National Science Foundation Graduate Research Fellowship in 2008. His research interests include robotics, system integration, human-robot collaboration and the commercialization of robotics technologies. After graduation, he will be working as a staff robotics engineer on the DARPA Robotics Challenge (DRC) project at the Carnegie Mellon University’s National Robotics Engineering Center (NREC) in Pittsburgh.
November 13, 2012
Dynamic software updating (DSU) systems allow programs to be updated while running, thereby permitting developers to add features and fix bugs without downtime. This paper introduces Kitsune, a new DSU system for C whose design has three notable features. First, Kitsune’s updating mechanism updates the whole program, not individual functions. This mechanism is more flexible than most prior approaches and places no restrictions on data representations or allowed compiler optimizations. Second, Kitsune makes the important aspects of updating explicit in the program text, making the program’s semantics easy to understand while minimizing programmer effort. Finally, the programmer can write simple specifications to direct Kitsune to generate code that traverses and transforms old-version state for use by new code; such state ransformation is often necessary, and is significantly more difficult in prior DSU systems. We have used Kitsune to update five popular, open-source, single- and multi-threaded programs, and find that few program changes are required to use Kitsune, and that it incurs essentially no performance overhead.
Speaker Biography: Michael W. Hicks is an associate professor in the Computer Science department and UMIACS at the University of Maryland, College Park, and is the Director of the Maryland Cybersecurity Center (MC2). His research focuses on using programming languages and analyses to improve the security, reliability, and availability of software. Noteworthy among his research accomplishments is the development of analysis and compilation tools for enabling software to be safely updated while it runs. He has explored the design of new programming languages and analysis tools for automatically discovering or remediating software flaws and security vulnerabilities. He has recently been exploring new approaches to privacy preserving computation and maintains an interest in distributed systems design and evaluation, particularly when adaptivity and security are system goals.
Distinguished Lecturer
November 15, 2012
A confluence of advances has led to an inflection in our ability to collect, store, and harness large amounts of data for generating insights and guiding decision making in the open world. Beyond study and refinement of principles, fielding real-world systems is critical for testing the sufficiency of algorithms and implications of assumptions-and exploring the human dimension of computational solutions and services. I will discuss several efforts pushing on the frontiers of machine learning and inference, highlighting key ideas in the context of projects in healthcare, transportation, and citizen science. Finally, I will describe directions with the composition of systems that draw upon a symphony of competencies and that operate over extended periods of time.
Speaker Biography: Eric Horvitz is a Distinguished Scientist at Microsoft Research. His interests span theoretical and practical challenges with developing systems that perceive, learn, and reason, with a focus on inference and decision making under uncertainty and limited resources. He has been elected a Fellow of the AAAI, the AAAS, and of the American Academy of Arts and Sciences. He received PhD and MD degrees at Stanford University. More information about his research, collaborations, and publications can be found at http://research.microsoft.com/~horvitz.
November 27, 2012
How does Alice convince Bob that she possesses a particular credential from Charlie? If Alice has a signature from Charlie on her public key, then all she will need to do is to show this signature to Bob, and also convince Bob that she is indeed Alice. In this setting, it is relatively clear how to revoke Alice’s credentials; also, one can limit the frequency and context of Alice’s credential use, and hold her accountable if she exceeds these limits.
How does Alice do this without revealing her identity to Bob, or indeed any identifying information? She produces a zero-knowledge proof of knowledge of the requisite values rather than showing them in the clear. Research on anonymous credentials concerns itself with protocols that enable Alice to obtain and demonstrate possession of credentials without revealing unnecessary information. It turns out that, similarly to the non-anonymous case, limits can be placed on the frequency and context of Alice’s use of anonymous credentials as well, and she can be identified and held accountable if she exceeds them.
Anonymous delegation is a research area that concerns itself with the next question: how does Alice delegate her credential to Bob without revealing any information about herself and learning anything about Bob?
In this talk, I will survey what we know so far about anonymous credentials, conditional anonymity and anonymous delegation. Specifically, I will outline generic constructions for these schemes, and give examples of specific instantiations. Some of these instantiations are efficient enough for practical use, have been implemented and piloted in real systems, and are part of a newly announced NIST-sponsored pilot on major U.S. university campuses in collaboration with Internet2.
This talk will be based on joint work with Mira Belenkiy, Jan Camenisch, Melissa Chase, Susan Hohenberger, Markulf Kohlweiss, Sarah Meiklejohn and Hovav Shacham.
Speaker Biography: Anna Lysyanskaya is an Associate Professor of Computer Science at Brown University. She received an A.B. in Computer Science and Mathematics from Smith College in 1997, and a Ph.D. in Computer Science and Electrical Engineering from MIT in 2002. She is a recipient of an NSF CAREER award and a Sloan Foundation fellowship and was included in the Technology Review Magazine’s list of 35 innovators under 35 for 2007. Her research interests are in cryptography, theoretical computer science, and computer security.
November 29, 2012
The analysis of structure in three-dimensional images is increasingly valuable for biomedical research and computational science. At the same time, the computational burden of processing images is increasing as devices produce images of higher resolution (e.g., typical CT scans have gone from 128^3 to roughly 512^3 resolutions). With the latest scanning technologies, it is also more common for the the values measured at each sample to be multi-dimensional rather than a single scalar, which further complicates implementing mathematically correct methods.
Diderot is a domain-specific language (DSL) for programming advanced 3D image visualization and analysis algorithms. These algorithms, such as volume rendering, fiber tractography, and particle systems, are naturally defined as computations over continuous tensor fields that are reconstructed from the discrete image data. Diderot combines a high-level mathematical programming notation based on tensor calculus with an abstract bulk-synchronous parallelism model. Diderot is designed to both enable rapid prototyping of new image analysis algorithms and high performance on a range of parallel platforms.
In this talk, I will give an overview of the design of Diderot and examples of its use. I will then describe aspects of its implementation.
Diderot is joint work with Gordon Kindlmann, Charisee Chiw, Lamont Samuels, and Nick Seltzer.
Speaker Biography: John Reppy is a Professor of Computer Science and a Senior Fellow of the Computation Institute at the University of Chicago. He received his Ph.D. from Cornell University in 1992 and spent the first eleven years of his career at Bell Labs in Murray Hill NJ. He has been exploring issues in language design and implementation since the late 1980’s, with a focus on higher-order, typed, functional languages. His work includes the invention of Concurrent ML and work on combining object-oriented and functional language features. His current research is on high-level languages for parallel programming, including the Manticore and Diderot projects.
Distinguished Lecturer
December 4, 2012
Haskell is a pure functional programming language, noted for features such as lazy evaluation, higher-order functions, a powerful type system (polymorphism, type classes, higher-order kinds), and computational abstractions (functors, monads, arrows). In this talk many of these features will be demonstrated in the context of computer music , both at the note level (representation, interpretation, algorithmic composition) and the signal level (audio processing, sound synthesis, instrument design). This will be a “show-by-example” talk – no knowledge of Haskell or music theory is assumed, but hopefully the examples will convey the general nature of Haskell and why it’s a great platform for computer music applications.
Speaker Biography: Paul Hudak is a Professor in the Department of Computer Science at Yale University. He has been on the Yale faculty since 1982, and was Chairman from 1999-2005. He received a BE from Vanderbilt University in 1973, an MS from MIT in 1974, and a PhD from the University of Utah in 1982. Professor Hudak helped to organize and chair the Haskell Committee, was co-Editor of the first Haskell Report in 1988, and has written a popular Haskell textbook. He has been a leader in the design of domain specific languages (embedded in Haskell) for a diverse set of applications, with a focus most recently on computer music and audio processing. With two of his colleagues, he designed the new Computing and the Arts major at Yale in 2009. Hudak was Editor-in-Chief of the Journal of Functional Programming and a founding member of IFIP Working Group 2.8 on Functional Programming. Among his honors, Professor Hudak is an ACM Fellow, and is a recipient of an IBM Faculty Development Award and an NSF Presidential Young Investigator Award. In 2009 he was appointed Master of Saybrook College at Yale University.
December 11, 2012
The Gaussian relay network is a natural candidate to model wireless networks. Unfortunately, it is not known how to determine the capacity of this model, except for very simple network topologies. For this reason, in 2007, Avestimehr, Diggavi, and Tse introduced a purely deterministic network model (the ADT model), which approximates the Gaussian relay network by capturing two key properties of wireless channels, namely broadcast and superposition. The ADT model attracted considerable attention because its capacity can be determined efficiently. Furthermore, as shown in 2009 by Amaudruz and Fragouli, also an optimal coding strategy for the ADT network can be found in polynomial time.
In this talk, I will show a strong connection between the ADT model and matroid optimization, which is a well-established field in combinatorial optimization. This connection allows us to exploit strong algorithmic and structural results from matroid optimization to optimize and analyze ADT networks. In particular, standard matroid optimization algorithms can be used to find an optimal coding strategy very efficiently, and structural results on matroids translate into interesting properties of the ADT model, like a max-flow min-cut theorem. The additional abstraction obtained by rephrasing the ADT model in the context of matroids furthermore allows us to tackle a variety of generalizations of the ADT model.
The goal of this talk is twofold. Apart from showing how to find optimal coding strategies for the ADT model, I want to give a quick introduction to the field of matroid optimization, which is an interesting algorithmic toolbox with numerous applications beyond network coding.
Based on joint work with Michel Goemans and Satoru Iwata.
Speaker Biography: Rico Zenklusen is an assistant professor in the Department of Applied Mathematics and Statistics at Johns Hopkins University. He received his PhD from ETH Zurich in 2008 and was a postdoc at EPFL and MIT before joining Johns Hopkins. His main research interests are in Combinatorial Optimization and its interfaces with Operations Research and Computer Science.
December 13, 2012
Computing is fundamentally intertwined with the use of language: how do we express ourselves through programs? Nowhere is this problem more important or less understood than in education – what languages are appropriate for middle school students? High school students? What domain should these languages be associated with?
This talk will address the issue of language and K-12 education. We will survey curriculum and languages currently used in this context and present experiences from the Western State Colorado University computer camp. This camp is a week long experience for middle school students in which we use game engine programming to explore a wide variety of topics in computer science, math, physics, and design. The goal or our camp is to use computing as a means to explore a wide range of STEM subject material in a creative way.
We have built a software environment based on Python, the Panda3D game engine, and reactive programming to enable our campers to create 3-D scenes and small games that illustrate fundamental ideas in computing, math, and physics. This language is based on Functional Reactive Programming (FRP), a framework for integrating time flow into a conventional programming language. I will discuss the semantics of our reactive system, demonstrate our software, show examples of student work from our camp, and address some of the high level issues associated with educational computing.
This talk is accessible to anyone interested in computing, math, or K12 education.
Speaker Biography: John Peterson is a professor at Western State Colorado University in Gunnison, Colorado where he comprises half of the Computer Science department there. He received his PhD from the University of Utah on 1984 and later worked at Yale with Paul Hudak on the Haskell language and Functional Reactive Programming. He is currently on sabbatical from Western and is hanging out with Scott Smith this semester.
Student Seminar
December 17, 2012
Ensuring data reliability has always been a key concern in large storage systems. There have been numerous schemes put forward to make data more durable, which include replication and/or erasure coding. Remote auditing can be used to ensure that data is available and free from corruption and/or deletion. In the first part of this dissertation, we outline how to make remote data auditing more robust against small targeted deletions. We then move on to the second part of the dissertation where we focus on the recovery problem, i.e. given that a (hardware or software) failure has occurred, the cost of recovering the data affected by that failure should be minimized. While there are lots of facets to such an open ended problem, we focus on the I/O aspect of it. We examine ways of minimizing the amount of I/O needed for recovery in the context of an erasure coded storage system. This involves analyzing the recovery characteristics of both existing codes, as well as introducing a new class of erasure codes which have been specifically designed for optimal recovery I/O performance. We provide an in-depth experimental evaluation of the theoretical results to show that reading a minimal set of symbols during recovery does indeed translate into practical savings in reconstruction I/O cost.
Speaker Biography: Osama Khan received the B. Sc. (Hons.) degree in Computer Science from Lahore University of Management Sciences, Lahore (Pakistan) in 2002, and proceeded to Germany in 2003 where he completed the Masters program offered by the University of Saarland, in collaboration with the German Center for Artificial Intelligence and Max-Planck Institut for Computer Science. He was then awarded the Fulbright Fellowship to pursue his Ph.D. at Johns Hopkins University in 2006. At JHU, he was a member of HSSL, which is headed by Dr. Randal Burns. His research focuses on data reliability and recovery issues in large storage systems. In the summer of 2012, he interned at Microsoft Research where he worked on the ThinCloud project, dealing with issues related to fault tolerance in a large distributed storage system. His research interests include replication, erasure coding, data auditing, cloud file systems and distributed storage systems.