Seminars: Spring 2016
Complete Dictionary Recovery over the Sphere
Speaker:John Wright, Columbia University
Time: 11:00 am - 12:00 pm Feb 4, 2016
Location: 2MTC, 10th floor, Room 10.099, Brooklyn, NY
We consider a complete dictionary recovery problem, in which we are given a data matrix Y, and the goal is to factor it into a product Y ~ A_0 X_0, with A_0 a square and invertible matrix and X_0 is a sparse matrix of coefficients. This is an abstraction of the dictionary learning problem, in which we try to learn a concise approximation to a given dataset. While dictionary learning is widely (and effectively!) used in signal processing and machine learning, relatively little is known about it in theory. Much of the difficulty owes to the fact that standard learning algorithms solve nonconvex problems, and are difficult to analyze globally.
We describe an efficient algorithm which provably learns representations in which the matrix X_0 has as many as O(n) nonzeros per column, under a suitable probability model for X_0. Previous efficient algorithms either only worked for very sparse instances (O(n^{1/2}) nonzeros per column) or required multiple rounds of SDP relaxation. Our results follow from a reformulation of the dictionary recovery problem as a nonconvex optimization over a high dimensional sphere. This particular nonconvex problem has a surprising property: once about n^3 data samples have been observed, with high probability the objective function has no spurious local minima.
This geometric phenomenon, in which seemingly challenging nonconvex problems can be solved globally by efficient iterative methods, also arises in problems such as tensor decomposition and phase recovery from magnitude measurements. We sketch these connections, and illustrate our results with applications in microscopy and computer vision.
Joint work with Ju Sun and Qing Qu (Columbia).
About the Speaker: John Wright is an Assistant Professor in the Electrical Engineering Department at Columbia University. He received his PhD in Electrical Engineering from the University of Illinois at Urbana-Champaign in 2009, and was with Microsoft Research from 2009-2011. His research is in the area of high-dimensional data analysis. In particular, his recent research has focused on developing algorithms for robustly recovering structured signal representations from incomplete and corrupted observations, and applying them to practical problems in imaging and vision. His work has received a number of awards and honors, including the 2009 Lemelson-Illinois Prize for Innovation for his work on face recognition, the 2009 UIUC Martin Award for Excellence in Graduate Research, and a 2008-2010 Microsoft Research Fellowship, and the Best Paper Award from the Conference on Learning Theory (COLT) in 2012, the 2015 PAMI TC Young Researcher Award, and the 2015 SPARS Best Student Paper Award (for PhD Advisees Ju Sun and Qing Qu).
Advanced Measurement and Testing Technologies for Smart Grid and Energy Internet
Speaker:Qi Huang, University of Electronic Science and Technology of China, China
Time: 11:00 am - 12:00 pm Feb 10, 2016
Location: 2MTC, 10th floor, Room 10.099, Brooklyn, NY
The development of smart grid revolutionarily changes the design, operation and control of power grid. The power grids of the future come into reality by enabling intelligent communication across sensing, measurement, and control layers of the existing power systems. Sensors and measurements become a core part of the grid and new challenging problems have to be dealt with and solved. Also, to fully enjoy the potential benefits of smart grid, advanced testing solution would have to be developed to verify the functionalities as well as performances. Improving the reliability and distribution of electricity through the use of testing & measurement equipment is critical for the growth of the smart grid.
Smart grid practice in China is quite different from rest of the world. And recently China starts an ambitious energy initiative ---- energy Internet. This talk will briefly present the smart grid development in China and introduce the energy Internet. The most updated technological development in measurement and testing solutions of power systems under the smart grid environment will also be discussed.
About the Speaker: Qi Huang was born in Guizhou Province, China. He received the B.S. degree in electricalengineering from Fuzhou University, Fuzhou, Fujian, China, in 1996, the M.S. degree in electrical engineering from Tsinghua University, Beijing, China, in 1999, and the Ph.D. degree in electrical engineering from Arizona State University, Tempe, in 2003.
Currently, he is a Professor at the University of Electronic Science and Technology of China (UESTC), Chengdu, Sichuan, China; Executive Dean of School of Energy Science and Engineering, UESTC; and the Director of Sichuan State Provincial Lab of Power System Wide-area Measurement and Control.
His current research and academic interests include power system instrumentation, power system monitoring and control, and integration of distributed generation into the existing power system infrastructure. Dr. Qi Huang has authored or co-authored more than 150 technical papers.
Designing and Learning Features for Music Information Retrieval
Speaker:Juan Pablo Bello, New York University
Time: 11:00 am - 12:00 pm Feb 11, 2016
Location: 2MTC, 10th floor, Room 10.099, Brooklyn, NY
This talk discusses a mix of concepts, problems and techniques at the crossroads of signal processing, machine learning and music. I will start by motivating the use of content-based methods for the analysis and retrieval of music. Then, I will introduce recent work done at my lab on a variety of music information retrieval (MIR) problems such as automatic chord recognition, music structure analysis, cover song identification and instrument recognition. In the process of doing so, I'll review the impact of feature design for specific MIR tasks, suggest that existing feature extraction methods in audio can be re-conceptualized as deep, multi-layer and trainable systems combining affine transforms and subsampling operations, and show a few examples where deep learning matches or outperforms the current state of the art in music and sound classification.
About the Speaker: Juan Pablo Bello is Associate Professor of Music Technology, and Electrical & Computer Engineering, at New York University, with a courtesy appointment at NYU's Center for Data Science. In 1998 he received a BEng in Electronics from the Universidad Simón Bolívar in Caracas, Venezuela, and in 2003 he earned a doctorate in Electronic Engineering at Queen Mary, University of London. Juan's expertise is in digital signal processing, computer audition and music information retrieval, topics that he teaches and in which he has published more than 70 papers and articles in books, journals and conference proceedings. In 2008, he co-founded the Music and Audio Research Lab (MARL), where he leads research on music informatics. His work has been supported by public and private institutions in Venezuela, the UK, and the US, including a CAREER award from the National Science Foundation and a Fulbright scholar grant for multidisciplinary studies in France. For a complete list of publications and other activities, please visit: https://wp.nyu.edu/jpbello/
Optimization for large-scale machine learning: big model and big data
Speaker:Anna Choromanska, New York University
Time: 11:00 am - 12:00 pm Feb 24, 2016
Location: Brooklyn, NY
The talk will focus on selected challenges in modern large-scale machine learning in two settings: i) big model (deep learning) setting and ii) big data setting. The first part of the talk focuses on the theoretical analysis of challenging non-convex learning setting: deep learning with multilayer networks. Despite the success of convex methods, deep learning methods, where the objective is inherently highly non-convex, have enjoyed a resurgence of interest in the last few years. Deep networks achieve state-of-the-art performances on a number of problems in image recognition, speech recognition, natural language processing, and video recognition, but they are poorly understood from the theoretical perspective. Recent advances in deep learning theory will be presented. The connection between the highly non-convex loss function of a simple model of the fully-connected feed-forward neural network and the Hamiltonian of the spherical spin-glass model will be established. It will be shown that under certain assumptions i) for large-size networks, most local minima are equivalent and yield similar performance on a test set, (ii) the probability of finding a “bad” local minimum, i.e. with high value of loss, is non-zero for small-size networks and decreases with network size, (iii) struggling to find the global minimum on the training set (as opposed to one of the many good local ones) is not useful in practice and may lead to overfitting. The advances made by this research in the field of deep learning and applications will be discussed. Modern machine learning approaches often use big models, like deep learning models discussed in the first part of the talk, to process and learn from the data. The recent widespread development of sensors, data-storage and data-acquisition devices has helped make big data-sets common place. The second part of the talk focuses on a big data setting and addresses the problem of scaling learning algorithms to big data. The multi-class classification problem will be addressed, where the number of classes is extremely large, with the goal of obtaining train and test time complexity logarithmic in the number of classes. A reduction of this problem to a set of binary classification problems organized in a tree structure will be discussed. A top-down online tree construction approach for constructing logarithmic depth trees will be demonstrated, which is based on a new objective function. Under favorable conditions, the new approach leads to logarithmic depth trees that have leaves with low label entropy. Discussed approach comes with theoretical guarantees following from convex analysis, though the underlying problem is inherently non-convex. General discussion concludes the talk.
About the Speaker: Anna Choromanska is a Post-Doctoral Associate in the Computer Science Department at Courant Institute of Mathematical Sciences, New York University. She is working in the Computational and Biological Learning Lab, which is a part of Computational Intelligence, Learning, Vision, and Robotics Lab, of prof. Yann LeCun. She graduated with her PhD from Columbia University, Department of Electrical Engineering, where she was the The Fu Foundation School of Engineering and Applied Science Presidential Fellowship holder. She was advised by prof. Tony Jebara. She completed her MSc with distinctions in the Department of Electronics and Information Technology, Warsaw University of Technology with double specialization, Electronics and Computer Engineering and Electronics and Informatics in Medicine. She was working with various industrial institutions, including AT&T Research Laboratories, IBM T.J. Watson Research Center and Microsoft Research New York. Her research interests are in machine learning, optimization and statistics with applications in biomedicine and neurobiology. She also holds a music degree from Mieczyslaw Karlowicz Music School in Warsaw, Department of Piano Play. She is an avid salsa dancer performing with the Ache Performance Group. Her other hobbies is painting and photography.
Acquiring Metric and Semantic Information using Autonomous Robots
Speaker:Nikolay A. Atanasov, University of Pennsylvania
Time: 11:00 am - 12:00 pm Feb 29, 2016
Location: 2MTC, 10th floor, Room 10.099, Brooklyn, NY
Abstract: Recent years have seen impressive progress in robot control and perception includingmanipulation, aggressive quadrotor maneuvers, dense metric map reconstruction, and object recognition in real time. The grand challenge in robotics today is to capitalize on these advances in order to enable autonomy at a higher-level of intelligence. It is compelling to envision teams of autonomous robots in environmental monitoring, precision agriculture, construction and structure inspection, security and surveillance, and search and rescue.
In this talk, I will emphasize that many such applications can be addressed by thinking about how to coordinate robots in order to extract useful information about the environment. More precisely, I will formulate a general active estimation problem that captures the common characteristics of the aforementioned scenarios. I will show how to manage the complexity of the problem over metric information spaces with respect to long planning horizons and large robot teams. These results lead to computationally scalable, non-myopic algorithms with quantified performance for problems such as distributed source seeking and active simultaneous localization and mapping (SLAM).
I will then focus on acquiring information using both metric and semantic observations (e.g., object recognition). In this context, there are several new challenges such as missed detections, false alarms, and unknown data association. To address them, I will model semantic observations via random sets and will discuss filtering using such models. A major contribution of our approach is in proving that the complexity of the problem is equivalent to computing the permanent of a suitable matrix. This enables us to develop and experimentally validate algorithms for semantic localization, mapping, and planning on mobile robots, Google's project Tango phone, and the KITTI visual odometry dataset.
About the Speaker: Nikolay A. Atanasov is a postdoctoral researcher at the department of Mechanical Engineering and Applied Mechanics at the University of Pennsylvania, Philadelphia, PA. He received a B.S. in Electrical Engineering from Trinity College, Hartford, CT, in 2008, and M.S. and Ph.D. degrees in Electrical and Systems Engineering from the University of Pennsylvania in 2012 and 2015, respectively. His research focuses on robotics, control theory, and computer vision and in particular on controlling teams of robots to collect metric and semantic information in applications such as environmental monitoring, security and surveillance, localization and mapping, search and rescue, and object recognition. His contributions were recognized by an award for the best Ph.D. dissertation in Electrical and Systems Engineering at the University of Pennsylvania in 2015.
Exploiting contact interactions for robust manipulation and locomotion skills.
Speaker:Ludovic Righetti, Max-Planck Institute for Intelligent Systems, Tübingen Germany
Time: 11:00 am - 12:00 pm Mar 2, 2016
Location: 2MTC, 10th floor, Room 10.099, Brooklyn, NY
What are the algorithmic principles that would allow a robot to run through a rocky terrain, lift a couch while reaching for an object that rolled under it or manipulate a screwdriver while balancing on top of a ladder? Our research tries to answer this seemingly naive question, which in fact resorts to understanding the fundamental principles of robotic locomotion and manipulation. One important aspect of our work focuses on the optimal exploitation of contact interactions between the robot and its environment to create more robust and efficient behaviors in uncertain and constantly changing environments. In particular, I will show recent results on legged robot control, where we use optimization techniques for realtime control of both contact interactions and robot motion. I will show experimental results on a torque controlled humanoid robot and recent trajectory optimization techniques to efficiently plan motion together with interaction forces during locomotion. The second part of the talk will focus on compliant manipulation. We developed complete systems capable of achieving complex autonomous manipulation tasks and I will show how the use of multimodal sensory information can be exploited to create very reactive behaviors and to learn contact interactions using machine learning techniques.
About the Speaker: Ludovic Righetti leads the Movement Generation and Control group at the Max-Planck Institute for Intelligent Systems (Tübingen, Germany) since September 2012 and holds a W2 group leader position since October 2015. Before, he was a postdoctoral fellow at the Computational Learning and Motor Control Lab (University of Southern California) between March 2009 and August 2012. He studied at the Ecole Polytechnique Fédérale de Lausanne (Switzerland) where he received a diploma in Computer Science (eq. MSc) in 2004 and a Doctorate in Science in 2008. He has received a few awards, most notably the 2010 Georges Giralt PhD Award given by the European Robotics Research Network (EURON) for the best robotics thesis in Europe, the 2011 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) Best Paper Award and the 2016 IEEE Robotics and Automation Society Early Career Award. His research focuses on the planning and control of movements for autonomous robots, with a special emphasis on legged locomotion and manipulation.
Website: http://motiongroup.is.tuebingen.mpg.de/
The Power of Nonconvex Paradigms for High-Dimensional Estimation
Speaker:Yuxin Chen, Stanford University
Time: 11:00 am - 12:00 pm Mar 3, 2016
Location: 2MTC, 10th floor, Room 10.099, Brooklyn, NY
Abstract: In various scenarios in the information sciences, one wishes to estimate a large numberparameters from highly incomplete / imperfect data samples. A growing body of recent work has demonstrated the effectiveness of convex relaxation --- in particular, semidefinite programming --- for solving many problems of this kind. However, the computational cost of such convex paradigms is often unsatisfactory, which limits applicability to large-dimensional data.
This talk follows another route: by formulating the problems into nonconvex programs, we attempt to optimize the nonconvex objectives directly. To illustrate the power of this strategy, we present two concrete stories. The first involves solving a random quadratic system of equations, which spans many applications ranging from the century-old phase retrieval problem to various latent-variable models in machine learning. The second is about recovering a collection of discrete variables from noisy pairwise difference measurements, which arises when one wishes to jointly align multiple images or to retrieve the genome phases from paired sequencing reads. We propose novel nonconvex paradigms for solving these two problems. In both cases, the proposed solutions can be accomplished within linear time, while achieving a statistical accuracy that is nearly un-improvable.
About the Speaker: Yuxin Chen is currently a postdoctoral scholar in the Department of Statistics at Stanford University, supervised by Prof. Emmanuel Candès. He received the Ph.D. degree in Electrical Engineering and M.S. in Statistics from Stanford University, M.S. in Electrical and Computer Engineering from the University of Texas at Austin, and B.E. in microelectronics from Tsinghua University. His research interests include high-dimensional structured estimation, convex and nonconvex optimization, statistical learning, and information theory.
Cyber-Physical System Design Using Contracts
Speaker:Pierluigi Nuzzo, UC Berkeley
Time: 11:00 am - 12:00 pm Mar 4, 2016
Location: LC400, 5 MetroTech Center, Brooklyn, NY
Abstract: The realization of complex cyber-physical systems is creating design and verification challenges that will soon become insurmountable with today’s engineering practices. While model-based design tools are already facilitating several design tasks, harnessing the complexity of the Internet-of-Things scenario is only deemed possible within a unifying methodology. This methodology should help interconnect different tools, possibly operating on different system representations, to enable scalable design space exploration and early detection of requirement inconsistencies.
In this talk, I show how a contract-based approach provides a formal foundation for a compositional and hierarchical methodology for cyber-physical system design, which can address the above challenges, and encompass both horizontal and vertical integration steps. I use assume-guarantee contracts and their algebra (e.g., composition, conjunction, and refinement) to support the entire design process and enable concurrent development of system architectures and control algorithms. In the methodology, the design is carried out as a sequence of refinement steps from a high-level specification to an implementation built out of a library of components at the lower level. Top-level system requirements are represented as contracts, by leveraging a set of formal languages, including mixed integer-linear constraints and temporal logic. Contracts are then refined by combining synthesis and optimization-based methods. I propose a set of optimization-based algorithms for efficient selection of cost-effective architectures under safety, reliability, and performance constraints over a large, mixed discrete-continuous design space. I demonstrate the effectiveness of the approach on industrial design examples, including aircraft electric power distribution and environmental control systems, showing, for instance, that optimal selection of industrial-scale power system architectures can be performed in a few minutes. Finally, I conclude by presenting future research directions towards a full-fledged integrated framework for system design.
About the Speaker: Pierluigi Nuzzo is a Postdoctoral Scholar at the Department of Electrical Engineering and Computer Sciences of the University of California, Berkeley. He received the Ph.D. in Electrical Engineering and Computer Sciences from the University of California at Berkeley in 2015. He also holds the Laurea (M.Sc.) degree in Electrical Engineering (summa cum laude) from the University of Pisa and the Sant'Anna School of Advanced Studies, Pisa, Italy. Before joining U.C. Berkeley, he was a Researcher at IMEC, Leuven, Belgium, and the University of Pisa, working on the design of energy-efficient A/D converters, frequency synthesizers for reconfigurable radio, and design methodologies for mixed-signal integrated circuits. His research interests include: methodologies and tools for cyber-physical system and mixed-signal system design; contracts, interfaces, and compositional methods for embedded system design; energy-efficient analog and mixed-signal circuit design. Pierluigi received First Place in the operational category and Best Overall Submission in the 2006 DAC/ISSCC Design Competition, a Marie Curie Fellowship from the European Union in 2006, the University of California at Berkeley EECS departmental fellowship in 2008, the U.C. Berkeley Outstanding Graduate Student Instructor Award in 2013, and the IBM Ph.D. Fellowship in 2012 and 2014.
Grid-Scale Energy Storage Integration in Power Systems: Methods & Case Studies
Speaker:Yury Dvorkin, University of Washington, Seattle
Time: 11:00 am - 12:00 pm Mar 7, 2016
Location: 2MTC, 10th floor, Room 10.099, Brooklyn, NY
As a result of large-scale integration of renewable generation, electric grids are being operated closer and closer to their technical limits. Lowering security margins demonstrate the urgent need for conceptual reassessment of operating and planning paradigms to timely and reliably meet ambitious state- and nation-wide renewable portfolio targets. This presentation will describe how stochastic renewable generation complicates power system operations and how these challenges can be dealt with by means of emerging energy storage technologies. Specifically, this presentation will focus on the application of mathematical optimization to optimal siting and sizing of grid-scale battery energy storage systems (BESSs) and their operations in power systems with renewables.
BESSs have been proven to be a technically feasible solution to improve utilization of renewable generation. However, the capital cost of such devices remains relatively, if not prohibitively, expensive. This naturally raises concerns over whether BESSs are an economically viable option and can be sustainably integrated in future power systems. This presentation will describe two bi-level models based on mixed-integer linear programming and computationally tractable solution strategies based on the duality approach that can be used to optimize energy storage siting and sizing decisions in a market environment. The use of these models makes it possible to simultaneously respect both the laws of physics (e.g., Kirchhoff’s law) and the laws of economics (e.g., the equilibrium of the supply and demand). The usefulness of the proposed method is demonstrated using numerical experiments carried out on the real-life ISO New England and Western Electricity Coordinating Council (WECC) systems.
Finally, this presentation concludes with policy recommendations derived from the simulations based on the proposed models and outlines future research plans that aim to assist decision-makers in achieving 2030 and 2050 renewable energy targets.
About the Speaker: Yury Dvorkin received the B.S.E.E. degree with the highest honors from Moscow Power Engineering Institute (Technical University), Moscow, Russia, in 2011. He is currently a Ph.D. candidate in electrical engineering at the Renewable Energy Analysis Laboratory (University of Washington, Seattle).
Previously, Yury was a graduate student researcher at Los Alamos National Laboratory’s Center for Nonlinear Studies (2014). His research interests include short- and long-term planning in power systems with renewable generation, power system economics and policy. Yury is a recipient of the Clean Energy Institute's Graduate Fellowship (2013–2014) and the Clean Energy Institute's Student Training & Exploration Grant (2014–2015). In 2014 and 2015 Yury was recognized as the Outstanding Reviewer by both the IEEE Transactions on Power Systems and IEEE Transactions on Sustainable Energy, the flagship journals in power system engineering.
Human-Centric Formal Methods: From Circuit to Cyber-Physical Systems
Speaker:Wenchao Li, SRI International, Menlo Park Silicon Valley
Time: 11:00 am - 12:00 pm Mar 8, 2016
Location: 2MTC, 10th floor, Room 10.099, Brooklyn, NY
Dependability is now a first-order concern for modern computing systems as they become omnipresent in our everyday lives. From embedded microcontrollers to self-driving cars, trustworthiness and safety are not only necessities but also much desired features. Achieving a high confidence in these properties requires rigorous analysis across many facets of the design process and is beyond the capabilities of current automation tools. Formal methods, in principle, can deliver provable guarantees, but the actual process from requirements to verification still makes heavy demands on the time, talent, and rigor of the users. In this talk, I argue that human inputs and insights are unlikely to be completely automated away. I will present my research and vision on human-centric formal methods, where formal analysis is used to enhance human understanding of designs or can incorporate human factors and capture design intents in principled ways.
In the first part of the talk, I will summarize my research on reverse engineering untrusted circuits. Reverse engineering helps to recover structures in designs and raise the level of abstraction so that the result is more human comprehensible. I will describe an equivalence-checking guided approach of deriving high-level structures from the bit-level descriptions of digital circuits, and demonstrate its effectiveness on designs several orders of magnitude larger than any prior approach, including an unknown test article from DARPA with half a million gates. In the second part of the talk, I will focus on the synthesis of provably correct human-in-the-loop controllers. Similar to semi-autonomous vehicles, these are controllers whose correctness depends not only on the autonomous controller, but also on the actions of the human operator as well as the interaction between the two entities. I will first give formal underpinnings to the problem and then describe how to automatically synthesize such controllers from logic specifications. I will conclude the talk by discussing my current research on the design and verification of distributed cyber-physical systems.
About the Speaker: Dr. Wenchao Li is currently a Computer Scientist at SRI International. He received his Ph.D. in Electrical Engineering and Computer Sciences from UC Berkeley in 2013. His research focuses on developing theory and tools for the construction of provably dependable systems. His research has been recognized with the ACM Outstanding Ph.D. Dissertation Award in Electronic Design Automation and the Leon O. Chua Award for outstanding achievement in nonlinear science.
Harnessing Flexible Transmission for Reliable and Economic Operation of Power Systems
Speaker:Mostafa Sahraei-Ardakani, Arizona State University
Time: 11:00 am - 12:00 pm Mar 9, 2016
Location: 2MTC, 10th floor, Room 10.099, Brooklyn, NY
The electric power system is evolving as renewable resources take a larger share in the global energy portfolio. Consequently, power system operation paradigms should evolve to support this transformation. Traditionally, system operators have exclusively used generation resources to serve the demand, at the lowest cost, while complying with the reliability standards. Although recent advancements have assisted inclusion of limited demand-side resources in power system operations, transmission flexibilities are yet to be utilized. Implementation of smart grid could allow operators to co-optimize flexible transmission alongside generation dispatch. Technologies that would allow such co-optimization include topology control, via transmission switching, and impedance control through variable-impedance power electronic devices. Flexible transmission provides significant power flow control, which can substantially increase the transfer capability over the existing network. This flexibility can be harnessed in various time stages, such as day-ahead and real-time markets, and for different applications, including but not limited to cost reduction and corrective adjustments in response to occurrence of contingencies. This talk will discuss the state of the art challenges of incorporating flexible transmission in power system operations and present results on a variety of large-scale systems. The results suggest that flexible transmission, as an essential ingredient of the smart grid, is a powerful solution to the challenging problems of future sustainable energy systems with high renewable penetration.
About the Speaker: Mostafa Ardakani received his Ph.D. in energy engineering from The Pennsylvania State University, University Park, PA in 2013. He also holds the M.S. and the B.S. degrees in electrical engineering from University of Tehran, Tehran, Iran. He is currently a post-doctoral scholar at Arizona State University, where he works on a variety of topics in power systems and electricity markets. He has the experience of working on an ARPA-E funded project on topology control, and an NSF/DHS funded project on cyber-physical security of the smart grid.
Development of Task Executive Robot for Disaster Rescue: DRC Trials and Finals 2015
Speaker:Kiwon Sohn, University of Nevada
Time: 11:00 am - 12:00 pm Mar 10, 2016
Location: 2MTC, 10th floor, Room 10.099, Brooklyn, NY
Letting workers to perform emergency operations within the first hours of the disaster is important for mitigation. However, it is too lethal for people to drive around toxic, contaminated or radioactive sites. A robot that drives can quickly reach the scene to perform rescue operations at the site, deliver logistics or even extract casualties. In 2012, the former DARPA Program Director Gill Pratt specifically mentioned that the Japan's Fukushima Daiichi accident is an example of a disaster that would have benefited from more capable robots. Today's humanoids however rarely have robust locomotion and balance control to overcome obstacles in disaster. As such, enabling robots to drive an unmodified vehicle, move quickly and handle various materials was identified by DARPA as a very important task.
Dr. Sohn’s central research objectives are to develop autonomous robot systems which can assist and replace human labors in various fields. His interdisciplinary research targets the following areas: 1) self-driving robots for disaster rescues and industry uses; 2) material-handling by full sized robot systems for various tasks. His research involves both a) computational modeling of robot system and b) test and verification processes with real-world applications such as DARPA Robotics Challenge (DRC). Core areas of his research include: 1) the trajectory optimization framework design (learning agent based) to plan and optimize humanoid motions under a variety of internal and external constraints; 2) development of prototype sensor head for humanoid’s vehicle driving and material handling task execution; 3) perception data processing and decision making in communication degraded environments. This research addresses how to combine inter-disciplinary technologies with complex robotic systems for uses in disasters and industries.
About the Speaker: Recently, Dr. Kiwon Sohn served as the Chief of Engineering in Team DRC-HUBO@UNLV for DRC-Finals 2015. Dr. Sohn was the team lead of research and development (R&D) and placed the team in Top 8 position against worldwide qualified finalist teams. Dr. Sohn also served as the lead researcher of the DRC-Trials 2013’s Track-A Team (DRC-HUBO). He led the development of humanoid mounting and driving with utility vehicle. Under NSF MRI-R2 program, Dr. Sohn had been the main care taker of 6 Hubo+ platforms in United States. He installed and managed robots of MIT, Purdue, OSU and Georgia Tech and trained students from 9 universities. Dr. Sohn received his Ph.D. in Mechanical Engineering from Drexel University. He received his M.S. and B.S. in Electrical Engineering from University of Pennsylvania and Kyungpook National University.
Silicon Integrated High-Density Electrocortical Interfaces
Speaker:Sohmyung Ha, University of California San Diego
Time: 11:00 am - 12:00 pm Mar 30, 2016
Location: 2MTC, 10th floor, Room 10.099, Brooklyn, NY
Recent interest and initiatives in brain research have driven a worldwide effort towards developing implantable neural interface systems with high spatiotemporal resolution and spatial coverage extending to the whole brain. Electrocorticography (ECoG) promises a minimally invasive, chronically implantable neural interface with resolution and spatial coverage capabilities that, when appropriately scaled, meet the needs of recently proposed brain initiatives. Current ECoG technologies, however, typically rely on cm-sized electrodes and wired operation, severely limiting their resolution and long-term use.
Our work has advanced micro-electrocorticography (µECoG) technologies for wireless high-density cortical neural interfaces in two main directions: flexible active µECoG arrays; and modular fully integrated µECoG systems. I will present a systematic design methodology which addresses unique design challenges posed by the extreme densities, form factors and power budgets of these fully implantable neural interface systems, with experimental validation of their performance for neural signal acquisition, stimulation, and wireless powering and data communication. Notable innovations include 1) first demonstration of simultaneous wireless power and data telemetry at 6.78 Mbps data rate over a single 13.56 MHz inductive link; 2) integrated recording from a flexible active electrode ECoG array with 85 dB dynamic range at 7.7 nJ energy per 16-b sample; and 3) the first fully integrated and encapsulated wireless neural-interface-on-chip microsystem for non-contact neural sensing and energy-replenishing adiabatic stimulation delivering 145 µA current at 6 V compliance within 2.25 mm³ volume.
About the Speaker: Sohmyung Ha received the B.S degree summa cum laude and the M.S. degree in electrical engineering from the Korea Advanced Institute of Science and Technology (KAIST), Daejeon, Korea, in 2004 and 2006, respectively. From 2006 to 2010, he worked as an analog and mixed-signal circuit designer at Samsung Electronics Inc., Yongin, Korea, where he was a part of the engineering team responsible for several of the world best-selling multimedia devices, smartphones and TVs. After an extended career in industry, he returned to academia joining the Department of Bioengineering, University of California San Diego, La Jolla, CA, USA, where he received the M.S. degree in bioengineering and is currently working toward the Ph.D. degree in bioengineering as a Fulbright Scholar.
His research aims at advancing the engineering and applications of silicon integrated technology interfacing with biology in a variety of forms ranging from implantable biomedical devices to unobtrusive wearable sensors. The engineering advances in the design of integrated circuits and system components target high performance, miniature form factor, and power autonomy in fully integrated interfaces. Targeted applications include self-powered biosensors for wearable health monitoring, subcutaneous glucose sensors for continuous monitoring in diabetes patients, scalable high-resolution retinal prostheses, and minimally invasive brain-computer interfaces for closed-loop remediation of neurological disorders such as epilepsy and Parkinson’s disease.
Learning and Control with Adaptive Dynamic Programming: From the Foundations to Cyber Physical Systems
Speaker:Haibo He, University of Rhode Island
Time: 11:00 am - 12:00 pm Mar 31, 2016
Location: 2MTC, 10th floor, Room 10.099, Brooklyn, NY
With the recent development of brain research and modern technologies, scientists and engineers will hopefully find efficient ways to develop brain-like intelligent systems that are highly robust, adaptive, scalable, and fault tolerant to uncertain and unstructured environments. Among many efforts toward this long-term objective, it is widely recognized that adaptive/approximate dynamic programming (ADP) could be a core methodology to accomplish the brain-like learning and optimization capabilities, and also to bring such a level of intelligence closer to reality through complex cyber physical systems (CPS) across different domains.
In this talk, I will introduce a new ADP framework with rich internal goal representation for improved learning and optimization capability over time. This architecture integrates an internal goal generator network to provide a more informative and detailed internal value representation to support the decision-making process. Compared to the existing ADP approaches with a manual or “hand-crafted” reinforcement signal design, our approach can automatically and adaptively develop the internal reinforcement signal over time, therefore improving the learning and control performance. This internal goal network can also be designed in a hierarchical way, to provide a multi-level value representation. I will present our recent development of numerous architectures and algorithms under this ADP framework, including heuristic dynamic programming (HDP), dual-heuristics dynamic programing (DHP), and globalized dual heuristic dynamic programing (GDHP). Furthermore, I will also present numerous applications including smart grid and human-robot interaction, to demonstrate its broader applications across a wide range of CPS. As a multi-disciplinary research topic, I will also briefly discuss the future research challenges and opportunities in this field.
About the Speaker: Haibo He is the Robert Haas Endowed Chair Professor and the Director of the Computational Intelligence and Self-Adaptive (CISA) Laboratory at the University of Rhode Island, Kingston, RI, USA. His primary research interests include adaptive dynamic programming, cyber physical systems, computational intelligence, cyber security, and various application domains. He has published one sole-author book, edited 1 book and 6 conference proceedings, and authored/co-authors over 200 peer-reviewed journal and conference papers, including several highly cited papers in IEEE Transactions on Neural Networks and IEEE Transactions on Knowledge and Data Engineering, Cover Page Highlighted paper in IEEE Transactions on Information Forensics and Security, and Best Readings of the IEEE Communications Society. He has delivered more than 40 invited talks around the globe. He was the Chair of IEEE Computational Intelligence Society (CIS) Emergent Technologies Technical Committee (ETTC) (2015) and the Chair of IEEE CIS Neural Networks Technical Committee (NNTC) (2013 and 2014). He served as the General Chair of 2014 IEEE Symposium Series on Computational Intelligence (IEEE SSCI’14, Orlando, Florida). He is currently the Editor-in-Chief of IEEE Transactions on Neural Networks and Learning Systems. He was a recipient of the IEEE International Conference on Communications (ICC) “Best Paper Award” (2014), IEEE CIS “Outstanding Early Career Award” (2014), National Science Foundation CAREER Award” (2011), and Providence Business News (PBN) “Rising Star Innovator” Award (2011).
Distributed Collaborative Area Coverage using Mobile Robots
Speaker:Anthony Tzes, University of Patras, Greece
Time: Apr 14, 2016
Location: Brooklyn, NY
The area coverage problem by a swarm of mobile robots is addressed in this talk. Each robot senses a not-necessarily omnidirectional (circular) area and its position is restricted within an a priori uncertainty regime. A distributed collaborative control scheme is proposed that can be implemented in a distributed manner, in the sense that each robot computes its control command using information from its neighbors through the minimization of a cost function. The domain of validity of each assigned function relies on the proper tessellation of the covered space. The space can be non-convex and the neighboring robots are not necessarily spatial (Delaunay) but should move in a manner that end-to-end connectivity is guaranteed using either RF or optical-signals. The proposed concept is presented for Unmanned Ground Vehicles and is extended to Unmanned Aerial Vehicles. Simulation and experimental studies are presented to highlight the efficiency of the suggested control scheme.
About the Speaker: Professor Anthony Tzes received his Ph.D. from Ohio State University in 1990. Soon after his tenure from NYU’s Polytechnic (now Tandon) School of Engineering, he moved at the Electrical and Computer Engineering Department of University of Patras in Greece, where he leads the Applied Networked Mechatronics Group. He was the past Department Head (2009-13) of this department and a current member of the University’s Board of Regents. His research interests are distributed collaborative control, navigation and control of UAVs, and medical robotics.
Liquid Cloud Storage: A new approach to large scale data storage
Speaker:Tom Richardson, Qualcomm Inc.
Time: 11:00 am - 12:00 pm Apr 14, 2016
Location: Brooklyn, NY
Large scale data storage is evolving as the amount and value of stored digital content continues to rapidly expand. Reliable distributed storage systems consist of hundreds to tens of thousands of potentially unreliable storage nodes. To protect the data from node failures current systems use replication or erasure codes that protect data objects by redundantly spreading them over a small number of nodes. We introduce a new solution that spreads data objects over a large number of nodes. Our solution offers exceptional object durability, minimizes storage overhead and repair traffic, and provides fast predictable access to data objects.
About the Speaker: Tom Richardson received his Ph.D. from the Laboratory for Information and Systems in MIT in 1990. From 1990-2000 he was a member of the Mathematical Research Center in Bell Labs, Lucent Technologies (originally AT&T Bell Labs). In 2000 he joined Flarion Technologies, which was spun out of Lucent and developed a pioneering IP based cellular system. In Flarion he held the title of V.P. and Chief Scientist and focused on the design and implementation of the physical layer. In 2006 Flarion was acquired by Qualcomm Inc., where he is currently employed with title V.P. Engineering. Dr. Richardson’s main research area is iterative coding systems. He is co-author, with Ruediger Urbanke, of a book on the subject entitled "Modern Coding Theory" and was co-winner of the 2002 and the 2013 Information Theory Paper award. He is a fellow of the IEEE, co-winner of the 2011 IEEE Koji Kobayashi award and the 2014 IEEE Hamming medal, and a member of the National Academy of Engineering.
Building mobile broadband coverage and quality maps using MONROE platform
Speaker:Ozgu Alay, Networks Department of Simula Research Laboratory, Norway
Time: 11:00 am - 12:00 pm Apr 15, 2016
Location: 2MTC, 9th floor, Room 9.101, Brooklyn, NY
Mobile broadband (MBB) networks underpin numerous vital operations of the modern society and are arguably becoming the most important piece of the modern communications infrastructure in the world. The use of MBB networks has exploded over the last few years due to the immense popularity of mobile devices such as smartphones and tablets, combined with the availability of high-capacity 3G/4G mobile networks. Given the increasing importance of MBB networks and the enormous expected growth in mobile traffic, there is a strong need for a better understanding of the fundamental characteristics of operational MBB networks and their relationship with the performance parameters of popular applications. This is crucial not only for improving the user’s experience for the services that are running on the current 3G/4G infrastructure, but also for providing feedback to the design of the upcoming 5G technologies. The first part of this talk will discuss how we addressed this need by building the first European transnational open platform, MONROE, for large-scale monitoring and assessment of MBB performance in heterogeneous environments. More specifically, the design choices and capabilities of the MONROE platform will be described. In the second part of this talk, the focus will be on a specific use case: coverage and quality maps for MBB networks. Given the increasing heterogeneity of technologies in the last mile of MBB networks, support for seamless connectivity across multiple network types relies on understanding the prevalent network coverage and quality profiles that can capture different available technologies in an area. This part of the talk will discuss how machine learning techniques can be applied to the data collected through MONROE platform to build coverage and quality maps, and how these maps can be leveraged in the design of context-aware solutions to improve the application performance and end-user experience.
About the Speaker: Dr. Özgü Alay received the B.S. and M.S. degrees in Electrical and Electronic Engineering from Middle East Technical University, Turkey, and Ph.D. degree in Electrical and Computer Engineering at Polytechnic Institute of New York University. After receiving her Ph.D., she has worked in broadcasting industry in Nevion, Norway as a software developer and as a product manager for the video gateways. Currently, she is a senior research scientist at Networks Department of Simula Research Laboratory, Norway. Her research interests lie in the areas of performance and reliability analysis of mobile broadband networks, multi-path transmission over heterogeneous networks and robust multimedia transmission over wireless networks. She is involved in many EU projects and currently coordinating the EU H2020 MONROE project.
Optical Coherence Tomography
Speaker:Joel S. Schuma, New York University
Time: 11:00 am - 12:00 pm Apr 21, 2016
Location: 2MTC, 10th floor, Room 10.099, Brooklyn, NY
Optical Coherence Tomography (OCT) has matured in less than twenty years from a single A-scan taking more than 20 minutes to perform to a technology capable of imaging at more than 300,000 A-scans per second. It is the most broadly and rapidly adopted technology ever used in ophthalmology. OCT software allows clinicians to measure neural tissue in great detail, including the optic nerve, peripapillary retinal nerve fiber layer (RNFL) and the macular ganglion cell complex (GCC). These tissues can be segmented from OCT images and analyzed in three dimensions. Further, the measurements can be compared to normative data, allowing the discrimination between health and disease (in this case, glaucoma) and assessment of the degree of glaucomatous abnormality. Clinicians now have available to them software for assessing statistically significant change over time, allowing the evaluation of glaucoma progression.
The question that arises is, “How does statistically significant glaucomatous progression as detected by OCT relate to clinically significant glaucomatous change?” There is no easy answer to this question. It is possible to equate degree of RNFL abnormality (structural glaucoma damage) with the amount of visual field abnormality (functional glaucoma damage). It is clear that a dead or absent neuron cannot function, and that when enough neurons are lost this must correspond to lost visual field. Several studies have shown that progressive loss of RNFL as measured by OCT corresponds to progressive visual field loss, and further, that OCT detects more change events in a given period of time than are detectable by conventional standard achromatic perimetry. The relationship between OCT measured RNFL thinning and visual fields should be interpretable in the same context that visual field loss relates to clinically significant change in visual function. This parallelism may permit the understanding of OCT identified structural change and clinically significant glaucomatous change.
About the Speaker: Joel S. Schuman, MD, FACS is Professor and Chairman of Ophthalmology, New York University School of Medicine. Formerly, he was the Eye and Ear Foundation Endowed Chair in Ophthalmology, Director of UPMC Eye Center, and Professor of Clinical and Translational Science and Professor of Bioengineering at the University of Pittsburgh. Dr. Schuman and his colleagues were the first to identify a molecular marker for human glaucoma, as published in Nature Medicine in 2001. He has been continuously funded by the National Eye Institute as a principal investigator since 1995, is principal investigator of a National Institutes of Health (NIH) grant to study novel glaucoma diagnostics, and is co-investigator of NIH grants for research into novel optical diagnostics and short pulse laser surgery and for advanced imaging in glaucoma. He is an inventor of optical coherence tomography (OCT), used world-wide for ocular diagnostics. Dr. Schuman has published more than 300 peer-reviewed scientific journal articles, has authored or edited 8 books, and has contributed more than 50 book chapters. In 2002 he received the Alcon Research Institute Award and the Lewis Rudin Glaucoma Prize, in 2006 the ARVO Translational Research Award, and in 2012 the Carnegie Science Center Award as well as sharing the Champalimaud Award (a 1 million Euro cash prize). In 2004, he was elected into the American Society for Clinical Investigation. In 2006 he received the Association for Research in Vision and Ophthalmology (ARVO) Translational Research Award. He was elected to the American Ophthalmological Society in 2008. In 2011 Dr. Schuman was the Clinician-Scientist Lecturer of the American Glaucoma Society. In 2013 he gave the Robert N. Shaffer Lecture at the American Academy of Ophthalmology (AAO) Annual Meeting, and received the AAO Lifetime Achievement Award. In 2014 he became a Gold Fellow of ARVO. He is named in Who’s Who in America, Who’s Who in Medical Sciences Education, America’s Top Doctors and Best Doctors in America, and has been named a Top Doctor by Pittsburgh Magazine each year since 2006.
Extending the lab’s reach: lab experiments in rural communities
Speaker:Andrew Bell, New York University
Time: 11:00 am - 12:00 pm Apr 28, 2016
Location: 2MTC, 10th floor, Room 10.099, Brooklyn, NY
Social data collection is hard. Social data collection in remote areas can be particularly hard, as costs, logistical issues, lack of laboratory controls, and lack of infrastructure restrict the kinds of data that can be collected. Pencils, paper, tokens, and cups are among the tools that have reliably provided for researchers for decades. However, these methods can be time intensive, prohibiting the study of complicated resource interactions or eliciting slow equilibria from groups whose attention and availability may be limited. Recent shifts in the cost of handheld computers have increased the appeal of computer-assisted interviewing and experiments. I present results and experiences from application of tablet and Android-based surveys and games in rural settings, in the hopes of highlighting current research possibilities and potential pitfalls.
About the Speaker: Andrew Reid Bell is Assistant Professor of Environmental Studies at New York University. Prior to joining NYU he was Research Fellow at the International Food Policy Research Institute (IFPRI) in Washington, DC, and Earth Institute Fellow at Columbia University. His work employs modeling tools and economic/behavioral experiments to examine resource behavior in rural contexts; current project areas include Malawi, Pakistan, Bangladesh, Cambodia, and Vietnam.
Research and Challenges of Multimedia Big Data
Speaker:Shu-Ching Chen, Florida International University
Time: 11:00 am - 12:00 pm Apr 29, 2016
Location: LC433, 5 MetroTech Center, Brooklyn, NY
The pervasiveness of mobile devices & consumer electronics and the popularity of Internet & social networks have generated huge amounts of multimedia information in various media types (such as text, image, video, and audio) shared among a large number of people. This creates the opportunities and intensifies the interest of the research community in developing methods to address multimedia big data challenges for real-world applications. Providing solutions to multimedia data such as images and videos brings about a higher level of difficulties at attempting to understand their semantic meaning. In this talk, I will discuss the research opportunities and challenges in multimedia big data, and introduce a coherent framework for multimedia big data management and retrieval, ranging from multimedia data processing to indexing, query, retrieval, and presentation. A set of core techniques, such as multimedia big data analysis, content-based image/video retrieval, and multimedia data mining, will be discussed in details and demonstrated using a prototype system. In addition, I will present the idea of applying these techniques to practical applications such as disaster information management.
About the Speaker: Dr. Shu-Ching Chen is an Eminent Scholar Chaired Professor in Computer Science in the School of Computing and Information Sciences (SCIS), Florida International University (FIU), Miami. He received his Ph.D. in Electrical and Computer Engineering from Purdue University, West Lafayette, IN, USA in 1998. He is the Director of Distributed Multimedia Information Systems Laboratory and Co-Director of the Integrated Computer Augmented Virtual Environment (I-CAVE) at SCIS. His main research interests include multimedia big data, content-based image/video retrieval, distributed multimedia database management systems, multimedia data mining, and disaster information management. Dr. Chen has authored and coauthored more than 300 research papers and four books. Dr. Chen has been the PI/Co-PI of many research grants from NSF, National Oceanic and Atmospheric Administration (NOAA), Department of Homeland Security, Army Research Office, Naval Research Laboratory (NRL), Florida Office of Insurance Regulation, IBM, and Florida Department of Transportation with a total amount of more than 26 millions.
Dr. Chen was named a 2011 recipient of the ACM Distinguished Scientist Award. He received the best paper award from 2006 IEEE International Symposium on Multimedia. He was awarded the IEEE Systems, Man, and Cybernetics (SMC) Society’s Outstanding Contribution Award in 2005 and was the co-recipient of the IEEE Most Active SMC Technical Committee Award in 2006. He was also awarded the Inaugural Excellence in Graduate Mentorship Award from FIU in 2006, the University Outstanding Faculty Research Award from FIU in 2004, the Excellence in Mentorship Award from SCIS in 2010, the Outstanding Faculty Service Award from SCIS in 2004 and 2014, and the Outstanding Faculty Research Award from SCIS in 2002 and 2012. He has been a General Chair and Program Chair for more than 55 conferences, symposiums, and workshops. He is the founding Editor-in-Chief of the International Journal of Multimedia Data Engineering and Management, and Associate Editor/Editorial Board of IEEE Multimedia, IEEE Trans. on Human-Machine Systems, and other 13 journals. He served as the Chair of IEEE Computer Society Technical Committee on Multimedia Computing. He is Co-Chair of IEEE Systems, Man, and Cybernetics Society’s Technical Committee on Knowledge Acquisition in Intelligent Systems. He was a steering committee member of the IEEE Transactions on Multimedia. He is a fellow of IEEE and SIRI.
Optical Signal Processing in CMOS-Compatible Silicon Nano-Photonics
Speaker:Mahmoud Rasras, Masdar Institute of Science and technology
Time: May 6, 2016
Location: LC433, 5 MetroTech Center, Brooklyn, NY
Developing CMOS silicon-based photonics holds the promise of disruptively creating new signal processing and transport solutions through the monolithic integration of electronic and photonic functions. Furthermore, integrating photonic components in a CMOS platform provides the benefits of a significant reduction in power consumption, cost, and size. In addition, there is a great potential in creating new functions from incorporating photonics and electronics on the same silicon chip. In this talk, implementation and integration challenges of such components in CMOS-compatible silicon will be presented. The focus will be placed on optical interconnects applications.
About the Speaker: Dr. Mahmoud Rasras is an Associate Professor in the Electrical Engineering and Computer Science (EECS) Department at Masdar Institute of Science and technology (MI). Prior to joining Masdar, he was a Member of Technical Staff at Bell Labs, Alcatel-Lucent, NJ, USA. He was working on the design of photonic components for next generation optical and microwave networks. His research spanned a wide range of activities including: microwave photonics, optical transceivers, photonics sensors, and all-optical logic technology for high-speed data security and encryption. Dr. Rasras received a PhD degree in physics from the Catholic University of Leuven, Belgium, where he conducted his research at IMEC (Interuniversity Microelectronics Center). His work was focused on developing a spectroscopic photon emission microscopy technique to study the reliability of VLSI semiconductor devices. Dr. Rasras published more than 80 journal and conference papers, and holds 32 issued US patents. He is also a Senior IEEE Member and an Associated Editor for Optics Express.
Optimized dielectric design of stator windings fed by fast front pulses
Speaker:Pablo Gomez, Western Michigan University
Time: 11:00 am - 12:00 pm May 10, 2016
Location: 2MTC, 10th floor, Room 10.099, Brooklyn, NY
AC rotating machines are exposed to fast-front pulses when fed by frequency converters for speed and torque control. This type of excitation produces dielectric stress in the machine’s insulation system which can result in premature deterioration or failure. In this presentation, the modeling approach followed to predict the transient voltage and electric field distribution in machine windings under fast-front pulse excitation will be presented. In addition, the application of different optimization algorithms to identify and evaluate modifications to current winding designs will be discussed, aiming at the minimization of dielectric stress. It will be shown that the optimized designs not only minimize such stress, but also result in a substantial reduction of transient overvoltages and the amount of insulation material required. Thus, the modification of current winding designs has the potential to increase the duty cycle of motors fed by frequency converters and at the same time reduce their manufacturing and maintenance costs.
About the Speaker: Dr. Pablo Gomez completed his B.S. degree in Mechanical and Electrical Engineering from Autonomous University of Coahuila, Mexico in 1999. He received the M.Sc. degree in Electrical Engineering from the Center for Research and Advanced Studies of the National Polytechnic Institute, Guadalajara, Mexico in 2002 and his Ph.D. in Electrical Engineering from the same institution in 2005. From August 2008 to July 2010 he was a postdoctoral fellow at NYU Tandon School of Engineering. From 2005 to 2014 he was a professor at the National Polytechnic Institute of Mexico, having previously worked in industry. Currently, he is an associate professor at the Electrical and Computer Engineering Department of Western Michigan University, Kalamazoo, MI. His research areas include modeling and design of power components, electromagnetic compatibility of HV systems, high voltage engineering and insulation systems. He is the author/co-author of over 50 technical publications including journal papers, international conference presentations and a book chapter.
On Good Lattices (and on the relation between lattices and codes)
Speaker:Ram Zamir, Tel Aviv University
Time: 2:00 pm - 3:00 pm May 19, 2016
Location: 2MTC, 10th floor, Room 10.099, Brooklyn, NY
A lattice is a discrete subgroup of the Euclidean N-dimensional space, which is closed under regular addition and reflection.
Lattices provide useful analytical and algorithmic tools for designing codes for digital communication. In particular, lattice codes are used for analog-to-digital conversion, a problem known also as "vector quantization" or "lossy source coding". In the dual "channel coding" problem, lattice codes combine coding and modulation for noise immunity in transmitting digital data over the additive white-Gaussian noise channel. More recently, lattice codes were found useful for multi-user communication, in setups such as coding with side information, interference cancellation, network coding, and more.
What is a good lattice? how do we know it exists? how can we construct it? The figure of merit depends, in fact, on the application. In the talk we shall assess lattice goodness for each of the communication problems above, and show the existence of asymptotically good lattices as the dimension N becomes large. We prove the existence result in two ways: first, indirectly, using the Minkowski-Hlawka theorem; second, by randomizing a linear-code based lattice construction ("construction A").
About the Speaker: Ram Zamir was born in Ramat-Gan, Israel in 1961. He received the B.Sc., M.Sc. (summa cum laude) and D.Sc. (with distinction) degrees from Tel-Aviv University, Israel, in 1983, 1991, and 1994, respectively, all in electrical engineering. In the years 1994 - 1996 he spent a post-doctoral period at Cornell University, Ithaca, NY, and at the University of California, Santa Barbara. In 2002 he spent a Sabbatical year at MIT, and in 2008 and 2009 short Sabbaticals at ETH and MIT. Since 1996 he has been with the department of Elect. Eng. - Systems at Tel Aviv University.
Ram Zamir has been consulting in the areas of radar and communications (DSL and WiFi), where he was involved with companies like Orckit and Actelis. During the period 2005-2014 he was the Chief Scientist of Celeno Communications. He has been teaching information theory, data compression, random processes, communications systems and communications circuits at Tel Aviv University. He is an IEEE fellow since 2010. He served as an Associate Editor for Source Coding in the IEEE transactions on Information Theory (2001-2003), headed the Information Theory Chapter of the Israeli IEEE society (2000-2005), and was a member of the BOG of the society (2013-2015). His research interests include information theory (in particular: lattice codes for multi-terminal problems), source coding, communications and statistical signal processing. His book "Lattice coding for signals and networks" was published in 2014.
A Total Variation model for Automatic Image Restoration
Speaker:Alessandro Lanza, University of Bologna, Italy
Time: 2:00 pm - 3:00 pm May 27, 2016
Location: 2MTC, 9th floor, Room 9.101, Brooklyn, NY
The popular Total Variation (TV) model for image restoration can be formulated as a Maximum A Posteriori (MAP) estimator which uses a half-Laplacian image-independent prior favoring sparse image gradients. We propose a generalization of the TV prior, referred to as TVp, based on a half-Generalized Gaussian Distribution (hGGD) with shape parameter p. An automatic estimation of p is introduced so that the prior better fits the real images' gradient distribution; we will show that, in general, the estimated p value does not necessarily require to be close to zero. The restored image is computed by using an Alternating Directions Methods of Multipliers (ADMM) procedure. In this context, a novel result in multivariate proximal calculus is presented which allows for the efficient solution of the proposed model. Numerical examples show that the proposed approach is particularly efficient and well suited for images characterized by a wide range of gradient distributions.
About the Speaker: Alessandro Lanza received the MS degree in Civil Engineering from the University of Pavia, Italy, in 2000 and the European PhD degree in Information Technology from the Advanced Research Centre on Electronic Systems for Information and Communication Technologies (ARCES), University of Bologna, in 2007. In 2006, he spent six months at the Ecole Polytechnique Federale de Lausanne (EPFL), Computer Vision Laboratory (CVLAB), as a doctoral fellow under the supervision of Prof. Pascal Fua. He is currently a Research Associate with the Department of Mathematics, University of Bologna. His research interests include numerical methods for inverse problems and image processing, computer vision, pattern recognition. His recent research work focuses on variational methods for image/signal restoration, with particular attention to non-smooth non-convex sparsity-promoting models for image denoising and deblurring. Alessandro Lanza also collaborates with Datalogic S.p.A., one of the world leader companies in the Automatic Data Capture and Industrial Automation field, where he deals with the development of mathematical and statistical methods for automatic barcode reading. He is a member of the INdAM-GNCS group.
Trainable iterative algorithms for computational sensing
Speaker:Ulugbek Kamilov, Mitsubishi Electric Research Laboratories (MERL)
Time: 11:00 am - 12:00 pm Jun 9, 2016
Location: 2MTC, 10th floor, Room 10.099, Brooklyn, NY
A powerful strategy for increasing the quality and resolution of medical, biological, and industrial images is to acquire larger quantities of data and to jointly reconstruct the complete signal by correctly integrating all the information available. The images are formed computationally with a very large-scale iterative optimization that maximally exploits suitable models for both the physical (forward model) and signal-related (regularization) aspects of the problem. While traditionally such models are designed manually and independently from the actual optimization algorithms, in this talk, we describe an alternative approach based on machine learning. Specifically, we propose to parametrize the model-related components of an iterative optimization algorithm, and then learn those parameters directly from a set of training data.
We will discuss two of our recent applications of the idea: (a) MSE optimal estimation of sparse signals from linear measurements; (b) optical tomographic imaging in the presence of multiple scattering. In both applications, we propose to interpret the iterations of the corresponding algorithms as an artificial multi-layer neural network, whose adaptable parameters correspond to the signal prior in (a) and to the voxel values of the 3D object in (b). Training the network to reproduce the data allows to form high-quality images. Our results suggest that such learning approaches can significantly boost the quality of imaging in a wide variety of applications.
About the Speaker: Ulugbek received his Ph.D. degree from the department of Electrical Engineering, École Polytechnique Fédérale de Lausanne (EPFL), Switzerland, in 2015.
Since 2015, he has been a Research Scientist with Mitsubishi Electric Research Laboratories (MERL), Cambridge, MA, USA. His research develops computational techniques for solving inverse problems for applications in biomedical or industrial imaging. Prior to joining MERL, Ulugbek was an Exchange Student with Carnegie Mellon University (CMU), Pittsburgh, PA, USA, in 2007, a Visiting Student at Massachusetts Institute of Technology (MIT), Cambridge, MA, USA, in 2010, and a Visiting Student Researcher at Stanford University, Stanford, CA, USA, in 2013. His research interests include statistical estimation, signal and image processing, computational imaging, and machine learning.
Optimal Control Problems with Time Delays
Speaker:Richard Vinter, Imperial College London, UK
Time: 11:00 am - 12:00 pm Jun 10, 2016
Location: 2MTC, 10th floor, Room 10.099, Brooklyn, NY
Control systems involving time delays are pervasive: they are encountered, for example, in process control, where they typically arise from transport delays associated with fluids flowing between reactors, in biological systems, in which delays are associated, for example, with the reproductive cycle, and remote control applications where we must take account of communications delays to ‘complete the loop’. This talk concerns optimal control problems involving time delay systems. It will provide an overview of analytical and computational tools available for solving such problems, from the earliest optimality conditions, in the form of a version of the Pontryagin Maximum Principle for time delay systems, to new optimality conditions for free end-time problems and related algorithms. The talk will also report on a case study concerning the solution of an optimal control problem with time delays, arising in the sustainable harvesting of a biological resource.
About the Speaker: Richard Vinter (Fellow of the IEEE and the UK Royal Academy of Engineering) is Professor of Control, Imperial College London. He obtained a PhD in Control Engineering at Cambridge University and subsequently was a postdoctoral fellow for two years in the Electronic Systems Laboratory MIT. He is former Head of the Control and Power Group in the EEE Department, and has served as a Dean in the Engineering Faculty, Imperial College.
His research is centered on optimal control, nonlinear systems, nonlinear filtering and differential games, with applications in power systems, aerospace and process control. He has published over 150 papers and two textbooks, including the monograph Optimal Control.
Biomedical Applications of High-frequency Ultrasound
Speaker:Jonathan Mamou & Jeffrey A. Ketterling, Lizzi Center for Biomedical Engineering at Riverside Research
Time: 11:00 am - 12:00 pm Jun 14, 2016
Location: 2MTC, 10th floor, Room 10.099, Brooklyn, NY
High-frequency ultrasound (HFU, >20 MHz) and very-high- frequency Ultrasound (VHFU, >200 MHz) offer means of investigating biological tissue at the microscopic level with spatial resolutions better than 100 µm and 10 µm, respectively. After a brief review of conventional ultrasound imaging, this talk will introduce HFU imaging and present a sampling of biomedical applications of HFU and VHFU.
One application of HFU is to acquire and form three-dimensional (3D) images of the brain of developing mouse embryos using annular-array transducer technology developed at Riverside Research. Many genetically-engineered mice display abnormal development of the central nervous system (CNS) at early embryonic stages and HFU offers great potential for in utero imaging and characterization. In utero, in vivo 3D data sets were acquired from 48 embryos using a 34-MHz, five-element annular array that was excited with a chirp-coded excitation. The embryos spanned the days E10.5 to E13.5, and volume renderings of the CNS were analyzed for changes related to growth. Along with imaging mouse embryos, the annular array can be used for ophthalmic imaging in humans. A real-time clinical prototype developed at Riverside is currently in use at Columbia University Medical Center. The development and performance of the prototype will be briefly discussed, and illustrative HFU images of the eye will be shown.
Another HFU application involves 3D imaging and characterization of freshly-excised human lymph nodes from cancer patients. Quantitative-ultrasound (QUS) images were formed and used to detect metastases using a 26-MHz center-frequency transducer. Classification results suggest that these methods may provide a new, clinically-important means of identifying small metastatic foci that might not be detected using standard pathology procedures. In a different HFU study, QUS approaches were also applied ex vivo to human cartilage for early detection and characterization of osteoarthritis, and initial results will be presented.
Finally, VHFU scanning-acoustic- microscopy images of rat-liver tissue, retina samples, and human lymph-node sections acquired using transducers ranging from 250-MHz to 500-MHz will be presented. Using 500-MHz VHFU, quantitative acoustic microscopy permits forming 2D images of speed of sound, acoustic impedance, and acoustic attenuation with spatial resolution better than 8 µm. These unique quantitative images can be used to measure tissue morphology, assess mechanical tissue properties, and better understand and model ultrasound scattering.
About the Speaker: Dr. Jonathan Mamou graduated in 2000 from the Ecole Nationale Supérieure des Télécommunications in Paris, France. In January 2001, he began his graduate studies in electrical and computer engineering at the University of Illinois at Urbana-Champaign, Urbana, IL. He received his M.S. and Ph.D. degrees in May 2002 and 2005, respectively. He is now Research Manager of the F. L. Lizzi Center for Biomedical Engineering at Riverside Research in New York, NY. His fields of interest include theoretical aspects of ultrasonic scattering, ultrasonic medical imaging, ultrasound contrast agents, and biomedical image processing. He is the co-editor of the book Quantitative Ultrasound in Soft Tissues published by Springer in 2013. Jonathan Mamou is a Senior Member of IEEE, a Fellow of the American Institute of Ultrasound in Medicine (AIUM), and a Member of the Acoustical Society of America. Dr. Mamou currently serves as the Chair of the AIUM High-Frequency Clinical and Preclinical Imaging Community of Practice. He is as an Associate Editor for Ultrasonic Imaging and the IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control and a reviewer for numerous journals. Dr. Mamou has authored or co-authored more than 100 peer-reviewed papers and conference proceedings, is an inventor on two patents and regularly serves on NIH study sections.
Dr. Ketterling joined the Lizzi Center for Biomedical Engineering at Riverside Research in 1999 as a Member of the Research Staff and, since 2014, has been the Associate Research Director for the group. He serves as the principal investigator for programs supported by the National Institutes of Health that deal with high-frequency annular arrays for small-animal and ophthalmic imaging, high-frequency acoustic contrast agents for microcirculation imaging in small-animals and eyes, and hydrophone arrays for characterizing the instantaneous acoustic fields of lithotripters. Dr. Ketterling was the Technical Chair for the Biomedical Acoustics Committee of the Acoustical Society of America (ASA) from 2008-2011 and he is a member of the IEEE International Ultrasonics Symposium Medical Ultrasonics Technical Program Committee. Dr. Ketterling is a Fellow of the ASA. He is also an Associate Editor for the journals Ultrasonic Imaging and IEEE-Transactions on Ultrasonics Ferroelectrics and Frequency Control. Dr. Ketterling has authored or co-authored more than 100 peer-reviewed papers and conference proceedings, is an inventor on five patents, acts as a peer reviewer for numerous acoustic-related journals and serves on NIH study sections. Dr. Ketterling received the B.S. degree in electrical engineering from the University of Washington, Seattle, WA, in 1994. He received the Ph.D. degree in mechanical engineering from Yale University, New Haven, CT, in 1999. His thesis concerned experimental studies of phase-space stability in single bubble sonoluminescence.
An Application, and a Standards-Driven Revisitation of Setuid
Speaker:Mahesh Tripunitara, University of Waterloo
Time: 11:00 am - 12:00 pm Jun 23, 2016
Location: 2MTC, 10th floor, Room 10.099, Brooklyn, NY
In this talk, I will discuss two pieces of my work, with students, related to setuid. Setuid is a POSIX standard for privilege-management.
In the first part of the talk, I will discuss work that adopts the semantics of the setuid bit for assessing security properties of stored-procedure authorization in a commercial database system. Our work has demonstrated that the system does not possess three desirable properties that we posed. I will discuss also the broader context: a systematic approach to testing real-world authorization systems for security properties.
In the second part of the talk, I will discuss work that revisits the setuid family from the standpoint of the POSIX standard. Specifically, I will address three questions. (1) Is the POSIX standard indeed broken, as prior work suggests? (2) Are implementations POSIX-compliant as claimed? (3) Are wrapper functions, that prior work proposes to circumvent issues, correct and usable?
This is joint work with my students Paul Bottinelli, Mark Dittmer and Alireza Sharifi.
About the Speaker: Mahesh Tripunitara is an Associate Professor in the ECE department at the University of Waterloo in Canada, where he had been since 2009. He works mostly in information security, on problems in access control, cryptographic key transport, the security of digital ICs and more recently, information leakage in cloud systems. He has a PhD in computer science from Purdue University, and about 9 years of industry-experience. His work with students has been awarded 'Best Paper' and 'Best Paper Runner-Up,' respectively, at the ACM Symposium on Access Control Models and Technologies (SACMAT) 2013 and 2015, and 'Best Student Paper' at the Usenix Security Symposium 2013.
Cooperative Control, Decision-Making, and Motion Planning for Human-Robot Collaboration Systems
Speaker:Yue “Sophie” Wang, Clemson University
Time: 11:00 am - 12:00 pm Jul 5, 2016
Location: 2MTC, 9th floor, Room 9.101, Brooklyn, NY
Autonomous systems are playing an ever more important role in every aspect of society. Human-robot collaboration integrates the best part of human intelligence with the advantages of autonomous systems. This talk will begin with a brief overview of the state-of-the-art in human-robot interaction. It will then present an overview of the current research projects at the Interdisciplinary & Intelligent Research (I2R) laboratory in the Mechanical Engineering Department at Clemson University in the area of cooperative control, decision-making, and motion planning for human-robot collaboration systems, including trust-based cooperative control, regret-based sequential decision-making, and high-level symbolic multi-robot motion planning with human-in-the-loop. Case studies will be presented in human-robot collaboration in assembly in manufacturing and teleautonomous operations of multi-agent systems. The talk will conclude with a brief presentation of the I2R lab facilities and experiment results.
About the Speaker: Dr. Yue “Sophie” Wang is an Assistant Professor of Mechanical Engineering and the Director of the I2R laboratory in the Mechanical Engineering Department at Clemson University. She received a Ph.D. degree in Mechanical Engineering from the Worcester Polytechnic Institute in 2011 and held a postdoctoral position in Electrical Engineering at the University of Notre Dame from 2011 to 2012. Her research interests are control of human-robot collaboration systems, symbolic robot motion planning, multi-agent systems, and cyber-physical systems. Dr. Wang has received an AFOSR YIP award in 2016, an NSF CAREER award in 2015, Air Force Summer Faculty Fellowship in 2015 and 2016, and the Clemson University Mechanical Engineering Eastman Chemical Award for Excellence in 2015. She is also the PI for Clemson's initiative on human-robot collaborative manufacturing, and was the PI for a NASA EPSCoR project from 2014 to 2015, and a Clemson University research grant from 2012 to 2013. Dr. Wang is a senior member of IEEE, and member of ASME and AIAA. She serves as the Co-Chair of the IEEE Control System Society Technical Committee on Manufacturing Automation and Robotic Control.
ADMM in Imaging Inverse Problems: Some History and Recent Advances
Speaker:Mário A. T. Figueiredo, Universidade de Lisboa, Portugal
Time: 11:00 am - 12:00 pm Jul 6, 2016
Location: 2MTC, 9th floor, Room 9.101, Brooklyn, NY
The alternating direction method of multipliers (ADMM) is an optimization tool of choice for several imaging inverse problems, namely due its flexibility, modularity, and efficiency. In this talk, I will begin by reviewing our early work on using ADMM to deal with classical problems such as deconvolution, inpainting, compressive imaging, and how we have exploited its flexibility to deal with different noise models, including Gaussian, Poissonian, and multiplicative, and with several types of regularizers (TV, frame-based analysis, synthesis, or combinations thereof). I will then describe more recent work on using ADMM for other problems, namely blind deconvolution and image segmentation, as well as very recent work where ADMM is used with plug-in learned denoisers to achieve state-of-the-art results in class-specific image deconvolution. Finally, on the theoretical front, I will describe very recent work on tackling the infamous problem of how to adjust the penalty parameter of ADMM.
About the Speaker: Mário A. T. Figueiredo received a PhD in electrical and computer engineering, from Instituto Superior Técnico (IST), the engineering school of the University of Lisbon, in 1994. He has been with the faculty of the Department of Electrical and Computer Engineering, IST, since 1994, where he is now a Professor. He is also area coordinator and group leader at Instituto de Telecomunicações, a private non-profit research institute. His research interests include image processing and analysis, machine learning, and optimization. M. Figueiredo is a Fellow of the IEEE and of the IAPR and is included in the 2014 and 2015 Thomson Reuters' Highly Cited Researchers lists; he received several awards, namely the 2011 IEEE Signal Processing Society Best Paper Award, the 2014 IEEE W. R. G. Baker Award, and the 2016 EURASIP Individual Technical Achievement Award for "fundamental contributions to optimization algorithms for sparsity-based and wavelet-based signal processing".