Seminars: Spring 2018
Speaker: Leandros Tassiulas, Yale University
Time: 11:00 am - 12:00 pm Feb 1, 2018
Location: 2 MetroTech Center, 10th floor, Room 10.099, Brooklyn, NY
Abstract: The virtualization of network resources provides unique flexibility in service provisioning in most levels of the network stack. Softwarization of the network control and operation (SDN) is a key enabler of that development. Starting from the network core, SDN is a dominant trend in the evolution of network architectures with increased emphasis recently on the network edge. I will present some recent results in this area starting with a study on migration from legacy networking to SDN enabled network modules. The trade-off between the benefits of SDN upgrades and the cost of deployment is addressed and captured by an appropriate sub-modular function that allows to optimize the penetration pace of the technology. Validation on some real world network topologies and traffic matrices will be presented as well. Then we move our attention to the network periphery. A wireless multi-hop extension at the network edge is considered and the problem of enabling SDN is addressed via replication of SDN controllers. The delay constraints of the controlled data-path elements is appropriately modeled and the problem of locating the controllers is addressed via optimization and a proof-of concept implementation. An alternate approach is considered then for the wireless network where we assume coexistence of SDN enabled components with network islands operating under distributed adhoc routing protocols. The trade-off of the coexistence is studied and the impact of SDN penetration is evaluated. Some paradigms of collaborative network services are presented finally as they are enabled by the above architectural evolution.
About the Speaker: Leandros Tassiulas is the John C. Malone Professor of Electrical Engineering at Yale University. His research interests are in the field of computer and communication networks with emphasis on fundamental mathematical models and algorithms of complex networks, architectures and protocols of wireless systems, sensor networks, novel internet architectures and experimental platforms for network research. His most notable contributions include the max-weight scheduling algorithm and the back-pressure network control policy, opportunistic scheduling in wireless, the maximum lifetime approach for wireless network energy management, and the consideration of joint access control and antenna transmission management in multiple antenna wireless systems. Dr. Tassiulas is a Fellow of IEEE (2007). His research has been recognized by several awards including the IEEE Koji Kobayashi computer and communications award (2016), the inaugural INFOCOM 2007 Achievement Award "for fundamental contributions to resource allocation in communication networks," several best paper awards including the INFOCOM 1994, 2017 and Mobihoc 2016, a National Science Foundation (NSF) Research Initiation Award (1992), an NSF CAREER Award (1995), an Office of Naval Research Young Investigator Award (1997) and a Bodossaki Foundation award (1999). He holds a Ph.D. in Electrical Engineering from the University of Maryland, College Park (1991). He has held faculty positions at Polytechnic University, New York, University of Maryland, College Park, University of Ioannina and University of Thessaly, Greece.
Speaker: Zhao Yuan, NYU Shanghai Presents Engineering
Time: 3:00 pm - 5:00 pm Feb 13, 2018
Location: 2 MetroTech Center, 10th floor, Room 10.099, Brooklyn, NY
Abstract: Optimal power flow (OPF) is the fundamental mathematical model to optimize the operation of power system. We propose second-order cone programming (SOCP) convex OPF models which are reformulated from the original nonconvex OPF model. The advantage of convex OPF over nonconvex OPF in terms of global optimality over local optimality is obvious. We then propose a modified Benders decomposition algorithm (M-BDA) to efficiently solve large-scale OPF model by decomposition and parallelization. The feasibility and optimality of the proposed M-BDA are analytically and numerically proved. Finally the applications of convex OPF in distribution locational marginal pricing, wind power integration and super grid coordinated energy dispatch are demonstrated.
About the Speaker: PhD Candidate, Erasmus Mundus Joint PhD Degree: KTH Royal Institute of Technology, Sweden Comillas Pontifical University, Spain Delft University of Technology, Netherlands.
Speaker: Maggie Cheng, NJIT
Time: 11:00 am - 12:00 pm Feb 15, 2018
Location: 2 MetroTech Center, 10th floor, Room 10.099, Brooklyn, NY
Abstract: An undetected topology change caused by overgrown trees is the root cause for the 2003 large scale blackout. The discrepancy between the actual topology of the power grid and the graph model used in state estimation is considered a topology error. Topology error was a reason that the WLS-based state estimation failed to converge and report alarms. This presentation will cover a non-WLS approach for topology error detection, and a learning framework to further identify the line outage. Time series analysis of multi-stream PMU data is used for feature extraction, and several statistical learning algorithms are trained based on these features. The line outage identification process is not tightly coupled with power flow analysis and state estimation as these tasks require detailed and accurate information about the system matrices, but can achieve a high detection rate comparable to the previous work that involves solving power flow and state estimation equations. Both single line outage and multiple simultaneous line outages are considered.
About the Speaker: Maggie Cheng is an associate professor in the Martin Tuckman School of Management in New Jersey Institute of Technology. She received a Ph.D. degree in Computer Science from the University of Minnesota at the Twin Cities in 2003. She was a faculty member in the department of Computer Science in Missouri University of Science and Technology before she joined NJIT in 2016. Her research focuses on data analytics methodology and optimization in complex network systems, and has a wide spectrum of topics including data analytics for cyber-physical systems, intrusion and attack detection in communication networks, user cyber behavioral analysis, network tomography, fault diagnosis, system level vulnerability analysis, as well as social network analysis on link formation, information propagation, and community detection. Her research has been sponsored by National Science Foundation, Department of Energy, Department of Education, Department of Transportation, and UM-System Research Board. She served as a guest editor of the journal of Combinatorial Optimization, associate editor of Nano Communication Networks Journal, and is currently on the editorial board of the International Journal of Sensor Networks. She served as an organizer and member of technical program committees in multiple international conferences.
Speaker: Yann LeCun, Facebook AI Research & New York University
Time: 10:00 am - 11:00 am Feb 20, 2018
Location: Pfizer Auditorium, 5 MetroTech Center, Brooklyn, NY, US
Abstract: Deep learning is causing revolutions in computer perception and natural language understanding. But almost all these successes largely rely on supervised learning, where the machine is required to predict human-provided annotations. For game AI, most systems use model-free reinforcement learning, which requires too many trials to be practical in the real world. But animals and humans seem to learn vast amounts of knowledge about how the world works through mere observation and occasional actions. Good predictive world models are an essential component of intelligent behavior: with them, one can predict outcomes and plan courses of actions. One could argue that prediction is the essence of intelligence. Good predictive models may be the basis of intuition, reasoning and "common sense", allowing us to fill in missing information: predicting the future from the past and present, the past from the present, or the state of the world from noisy percepts. After a brief presentation of the state of the art in deep learning, some promising principles and methods for predictive learning will be discussed.
Speaker: Brendan Englot, Stevens Institute of Technology
Time: 11:00 am - 12:00 pm Feb 23, 2018
Location: 2 MetroTech Center, 9th floor, Room 9.011, Brooklyn, NY
Abstract: This seminar will describe recent research efforts to produce lightweight and robust autonomous navigation solutions for range-sensing mobile robots at opposite ends of the spectrum: (1) unmanned ground vehicles (UGVs) equipped with dense, volumetric scanning lidar, and (2) underwater remotely operated vehicles (ROVs) supported by sparse, noisy sonar data. In both cases, these robots must efficiently and safely navigate complex, unstructured environments in the presence of clutter. For range-sensing UGVs, I will discuss efficient techniques for perception and decision-making over dense and voluminous data. This includes a navigation package that relies solely on a Velodyne "Puck" lidar and an Nvidia Jetson TX2 embedded computer to achieve (1) computationally lightweight, low-drift lidar odometry, and (2) real-time scan-by-scan terrain traversability mapping. These techniques in turn support learning-enabled decision-making that guides efficient autonomous exploration of unknown environments. For sonar-equipped ROVs, I will discuss techniques that leverage probabilistic inference to support decision-making over sparse and noisy data. This includes a pipeline for accurate localization and mapping with small quantities of features, occupancy mapping that leverages probabilistic inference to produce predictive 3D grid maps of submerged structures, and an active-SLAM exploration algorithm that uses virtual landmarks to produce improved-accuracy maps. Finally, in support of robust underwater maneuvering and control, I will discuss approaches for efficient motion planning under localization uncertainty, and for robot reinforcement learning under heteroscedastic noise.
About the Speaker: Dr. Brendan Englot received S.B., S.M. and Ph.D. degrees in Mechanical Engineering from the Massachusetts Institute of Technology in 2007, 2009 and 2012, respectively. At MIT, he studied motion planning for surveillance and inspection applications, deploying his algorithms on an underwater inspection robot that is now being produced in quantity for the US Navy. During 2012-2014, Brendan was with United Technologies Research Center in East Hartford, Connecticut, where he was a Research Scientist and Principal Investigator in the Autonomous and Intelligent Robotics Laboratory (AIRLab) and a technical contributor to the Sikorsky Autonomous Research Aircraft (SARA). At Stevens Institute of Technology, he directs the Robust Field Autonomy Lab, which focuses on robust autonomous navigation solutions for robots operating in harsh and unstructured environments. Brendan is the recipient of a 2017 NSF CAREER award.
Speaker: Franziska Meier, Max-Planck Institute for Intelligent Systems
Time: 11:00 am - 12:00 pm Feb 26, 2018
Location: 6 MetroTech Center, JAB674, Brooklyn, NY
Abstract: With recent advances in machine learning for intelligent agent design, the robotics community has made immense progress towards autonomous skill learning. However, the day of deploying robots to assist people in their homes still seems far away. One of the key challenges towards this goal is to enable a robot to robustly and safely deal with unknown situations and unexpected events.
Current robot learning approaches assume that one can prepare a robot for every possible task and environment variation. With this assumption learning becomes a large data collection effort, followed by a one-time training phase. Once training has converged, learning of the system is terminated and the robot is left to execute it’s skills in the real world, without the ability to further adapt and improve them. Yet, to be truly autonomous, robots need to be able to react to unexpected events and then update their models to include the just encountered data points. Furthermore, a robot’s ability to adapt to new situations should improve over time. In short, true autonomy requires continuous learning. However, continuously updating models without forgetting previously learned mappings remains a fundamental open research problem. In this talk, I will present learning algorithms, based on localized inference schemes, that alleviate the problem of forgetting when learning continuously. Furthermore, I will introduce our recent advances on learning-to-learn for robotics, which accelerates learning of novel task variations. I will demonstrate the effectiveness of our learning algorithms on the challenging task of learning the dynamics of a 7-DOF torque-controlled manipulator, which operates at a 1000Hz. Finally, I will conclude this talk by presenting my vision on how to enable life-long learning for robotics.
About the Speaker: Franziska Meier is a research scientist at the Max-Planck Institute for Intelligent Systems since August 2016. She is also a postdoctoral researcher with Dieter Fox at the University of Washington, Seattle, since February 2017. Before that she was a PhD student at the University of Southern California. She defended her thesis on “Probabilistic Machine Learning for Robotics” in 2016, under the supervision of Prof. Stefan Schaal. Prior to her PhD studies, she received her Diploma in Computer Science from the Technical University of Munich and attended the Georgia Institute of Technology as graduate student in Computer Science. Her research focuses on machine learning for robotics, with a special emphasis on continuous learning for robotics.
Speaker: Giuseppe Loianno, University of Pennsylvania
Time: 11:00 am - 12:00 pm Mar 2, 2018
Location: 2 MetroTech Center, 9.011, Brooklyn, NY
Abstract: Flying robots are starting to play a major role in several tasks such as search and rescue, interaction with the environment, inspection, patrolling and monitoring. Unfortunately, their dynamics make them extremely difficult to control and this is particularly true in absence of external positioning systems, such as GPS and motion-capture systems. Additionally, autonomous maneuvers based on onboard sensors are still very slow compared to those attainable with motion capture systems. Agile navigation of Micro Aerial Vehicles (MAVs) through unknown environments poses a number of challenges in terms of perception, state estimation, planning, and control. To achieve this, MAVs have to localize themselves and coordinate between each other in unstructured environments. This in turn requires the MAV to use a combination of absolute or relative asynchronous measurements provided by different noisy sensors at different rates which have to be fused to obtain a reliable state estimate at rates of above 200 Hz. In this talk, I will present recent research results on the pose estimation and planning problems for agile flights, transportation and physical interaction with the environment using a minimal onboard sensor suite composed mainly by a single camera system and an Inertial Measurement Unit (IMU). For truly autonomous agile navigation, the perception, planning and control problems have to be tightly integrated and solved concurrently. I will demonstrate how these different technologies can be combined to design new on-board algorithms, which allow a robust solution enabling aggressive and agile flight maneuvers with MAVs in different scenarios including the ability to interact with the environment.
About the Speaker: Dr. Giuseppe Loianno is a lecturer, research scientist, and team leader at the General Robotics, Automation, Sensing and Perception (GRASP) Laboratory at the University of Pennsylvania. He received his BSc and MSc degrees in automation engineering, both with honors, from the University of Naples "Federico II" in December 2007 and February 2010, respectively. He received his PhD in computer and control engineering focusing in robotics in May 2014 in the PRISMA Lab Group, led by Prof. Dr. Bruno Siciliano. He has been involved in the EU FP7 project AIRobots (www.airobots.eu) in sensor fusion and visual control. Dr. Loianno has published more than 40 conference papers, journal papers, and book chapters. From April 2013 he worked for 14 months with the GRASP Lab at the University of Pennsylvania, supervised by Prof. Dr. Vijay Kumar. From June 2014 to July 2015, he was a postdoctoral researcher in his lab, where he is currently a research scientist, lecturer and team leader. His research interests include visual odometry, sensor fusion, and visual servoing for micro aerial vehicles. He is worldwide recognized for his expertise in autonomy for small aircraft. He received the Conference Editorial Board Best Reviewer Award at ICRA 2016. His work has been featured in a large number of renowned news and magazines.
Speaker: Paul Pearce, University of California, Berkeley
Time: 11:00 am - 12:00 pm Mar 5, 2018
Location: 5 MetroTech Center, LC400, Brooklyn, NY
Abstract: The value and power mediated by the global, interconnected systems of today’s Internet attract adversaries who seek to exploit these systems for economic, political or social gain. Yet, the underlying complexity of Internet infrastructure, the layering of its services, and the indirect nature of its business relationships can make it challenging to identify even the existence of adversaries manipulating systems for their benefit.
In this talk I present systems and methods to uncover and explore two such large-scale adversarial activities, Internet-wide cybercrime and censorship. I begin with an in-depth exploration of ZeroAccess, a complex peer-to-peer botnet which served as a delivery platform for advertising abuse malware for more than four years and impacted millions of users. I identify innovative attacks and fraudulent business relationships within the advertising ecosystem stemming from complex multi-hop ad reseller chains, resulting in millions of dollars in fraud per month. These relationships and explorations were used as a focal point for fraud remediation and a takedown of the botnet.
Next I present Augur, a measurement technique and accompanying system that uses highly noisy TCP/IP side channels to measure reachability between two Internet locations without access to the endpoints or the path between them. Augur uses sequential hypothesis testing to provide statistical confidence in the face of network and side channel noise. I then use Augur to perform a global censorship measurement study of the blocking practices of more than 180 countries.
About the Speaker: Paul Pearce is a PhD Candidate at UC Berkeley advised by Vern Paxson and a member of the Center for Evidence-based Security Research (CESR). By developing Internet-scale measurement platforms and new empirical methods, his research brings grounding and understanding to the study of large-scale, hidden Internet security problems. His work spans the areas of cybercrime, censorship, and "advanced persistent threats" (APTs). His work has been distinguished at the IEEE Symposium on Security and Privacy, and he has been recognized as an EECS Distinguished Graduate Student Instructor.
Speaker: William J. Beksi, University of Minnesota, Twin Cities Department of Computer Science and Engineering
Time: 11:00 am - 12:00 pm Feb 27, 2018
Location: 2 MetroTech Center, Room 9.009, Brooklyn, NY
Abstract: 3D point cloud datasets are becoming more common due to the availability of low-cost sensors. Light detection and ranging (LIDAR), stereo, structured light, and time-of-flight are examples of sensors that capture a 3D representation of the environment. These sensors are increasingly found in mobile devices and machines such as smartphones, tablets, robots, and autonomous vehicles. This talk will cover novel techniques for generating, processing, and distributing 3D point cloud data for robotic perception tasks. First, topological methods for 3D hole boundary point detection, segmentation, and object detection and classification will be introduced. Next, a 3D perception architecture for robots connected to a remote computing infrastructure will be shown.
Then, a framework to generate synthetic 3D scenes for the purpose of training data-driven machine learning methods such as deep learning will be presented. Finally, the talk will conclude with a future roadmap for each of these emerging technologies.
About the Speaker: William J. Beksi is a PhD candidate in the Department of Computer Science and Engineering at the University of Minnesota. He is a research assistant at the Center of Distributed Robotics led by Professor Nikolaos Papanikolopoulos. His research focuses on developing algorithms and data structures for fundamental robotic perception tasks such as 3D reconstruction, segmentation, and object detection and classification.
He also works on problems in networked and cloud robotics where the goal is to allow under-resourced robots remote access to computing, data, and learning resources. During the past summer at iRobot, William spearheaded a project on synthetic 3D scene generation for training data-driven machine learning methods. His work has been published in top-tier robotics and computer vision venues. William is also the recipient of the MnDRIVE PhD fellowship through the University of Minnesota Informatics Institute for his research on topological methods for 3D point cloud processing.
Design, Development, and Evaluation of a Meso-Scale Robotic System for Deep Intracranial Tumor Removal
Speaker: Shing Shin Cheng, Georgia Institute of Technology
Time: 11:00 am - 12:00 pm Mar 6, 2018
Location: 2 MetroTech Center, 9.009, Brooklyn, NY
Abstract: Most of the existing surgical robotic systems have bulky footprint, employ straight rigid surgical tools, and are not MRI-compatible. The Minimally Invasive Neurosurgical Intracranial Robot (MINIR-II) project aims at combining the flexible robotic technology, minimally invasive approach, rapid prototyping, and MRI imaging modality to achieve more precise and complete removal of brain tumor. MINIR-II is a spring-based 3-D printed flexible robot that is tendon-driven and equipped with electrocautery, suction and irrigation capabilities. A novel central tendon routing mechanism has been employed to enable independent segment control of the multi-DoF robot in a tight workspace. To improve stability of the surgical procedure, a stiffness-tunable MINIR-II with shape memory alloy (SMA) spring segments has also been developed and characterized to investigate the effect of tendon locking and SMA segment stiffening on the stiffness of the individual segment. SMA springs have been used as the initial proof-of-concept MRI-compatible actuator for MINIR-II. To improve the actuation bandwidth of the SMA actuators, which is part of our effort to confirm the motion capability of the robot, cooling module-integrated SMA springs have been developed together with a new actuation mechanism involving the alternate passage of water and compressed air. In our most recent work, ultrasonic motors have replaced the SMA springs to provide a reliable remote actuation solution that improves the MR image quality. A complete set of MRI-compatible transmission mechanisms, including a switching mechanism, a linkage mechanism, and a quick-connect mechanism, has been developed. The robotic system is currently being evaluated in terms of its precision, force output, repeatability, and effect on the MRI signal-to-noise-ratio (SNR). Our work hopefully lays a solid foundation towards the future development of more MRI-compatible surgical robotic systems that deploy miniaturize, multi-DoF, meso-scale flexible robots, that are potentially customizable.
About the Speaker: Shing Shin Cheng received his B.S. degree in Mechanical Engineering from the Johns Hopkins University in 2013; M.S. degree in Mechanical Engineering from University of Maryland, College Park in 2016; and is currently a Ph.D. candidate in Robotics at Georgia Institute of Technology, Atlanta. His research interests include design and control of flexible surgical robots and medical devices, as well as exploration of smart materials, microfabrication, and artificial intelligence for medical applications.
Speaker: Arsalan Mosenia, Princeton University
Time: 11:00 am - 12:00 pm Mar 26, 2018
Location: 6 MetroTech, JAB674, Brooklyn, NY
Abstract: Internet of Things (IoT) is envisioned as a holistic and transformative approach for providing numerous services. Smart things, that can sense, store, and process electrical, thermal, optical, chemical, and other signals to extract user- /environment-related information, have enabled services only limited by human imagination. Despite picturesque promises of IoT-enabled systems, the integration of smart things into the standard Internet introduces several security and privacy challenges because the majority of Internet technologies, communication protocols, and sensors were not designed to support IoT. In this presentation, I will shed light on fundamental security challenges in IoT paradigm and argue that we need to rethink the development of multiple IoT-enabled systems while taking security requirements into account. Bridging concepts from information security, machine learning, and network science, I will demonstrate that the threat of unintended private information leakage from seemingly non-critical data is far beyond what is currently thought possible. In particular, I will describe PinMe, a novel user- location mechanism that exploits non-sensory/sensory data collected from smartphones or Internet-connected vehicles, along with publicly-available auxiliary information, e.g., elevation maps, to estimate the user's location when all location services, e.g., Global Positioning System (GPS), are turned off. Next, I will present a novel framework that integrates programmability, connectivity, and security into isolated vehicles and enables rapid development of new vehicular applications for already-in- market vehicles, significantly enhancing the vehicle security, passenger safety, and driving experience. The proposed framework is formed around a security/privacy-friendly programmable dongle (known as SmartCore) and a middlware that enables developers to interact with the vehicle's built-in components in a safe and secure manner, preventing numerous potential threats against Internet-connected vehicles.
About the Speaker: Arsalan Mosenia is currently a postdoctoral research associate, jointly working with Profs. Mung Chiang (Purdue University) and Prateek Mittal (Princeton University). He received the B.Sc. degree in Computer Engineering from Sharif University of Technology in 2012, and the M.A. and Ph.D. in Electrical Engineering from Princeton University, in 2014 and 2016, respectively, under the supervision of Prof. Niraj K. Jha. He is broadly interested in investigating and addressing emerging security and privacy challenges in Internet of Things (IoT) and cyber-physical systems. His interests lie at the intersection of information security, IoT, embedded systems, and machine learning. His work has uncovered fundamental security/privacy flaws in the design of multiple widely- used Internet-connected systems. His research impact includes several publications that are among the most popular papers of top-tier IEEE Transactions, multiple prestigious awards (including Princeton X, Princeton Innovation Fund, French-American Doctoral Exchange Fellowship, and Princeton IP Accelerator Fund), and extensive press coverage. Furthermore, at OpenFog Consortium, he is actively collaborating with Security Work Group, where he defines domain-specific security standards for fog computing, and Testbed Work Group, where he designs, builds, and examines novel fog-inspired real-world systems.
Speaker: Qi Alfred Chen, University of Michigan, Ann Arbor
Time: 11:00 am - 12:00 pm Mar 20, 2018
Location: 5 MetroTech Center, LC400, Brooklyn, NY
Abstract:The world is increasingly connected through a series of smart, connected systems such as smartphone systems, smart home systems, and the emerging smart transportation and autonomous vehicle systems. While leading to improved services, such transformation also introduces new security challenges. To address these challenges, in contrast to existing defense mechanisms that are mostly ad hoc and reactive, my research aims at developing proactive defense approaches that can systematically discover, analyze, and mitigate new security problems in smart, connected systems.
In this talk, I will focus on my research efforts in securing two most basic components in any smart, connected system: network stack and smart control. For network stack security, I will describe our discovery of a new attack vector (US-CERT alert TA16-144A) that was unexpectedly brought by the recent expansion in DNS, and our subsequent systematic analysis at both network and software levels for its defense. For smart control security, I will describe my most recent work that performed the first security analysis of the next-generation Connected Vehicle (CV) based traffic signal control, which discovers new vulnerabilities at the traffic signal control algorithm level. I will conclude by discussing my future research plans in securing existing and future smart, connected systems, especially those in critical domains such as transportation and automobile.
About the Speaker: Qi Alfred Chen is a Ph.D. candidate in the EECS department at University of Michigan advised by Professor Z. Morley Mao. His research interest is network and systems security, and the major theme of his research is to address security challenges through systematic problem analysis and mitigation. His research has discovered and mitigated security problems in various systems such as next-generation transportation systems, smartphone OSes, network protocols, DNS, GUI systems, and access control systems. His work has impact in both academia and industry with over 10 top-tier conference papers, news coverage and interviews, vulnerability disclosures, and industry discussions and responses. His current research focuses on smart systems and IoT, e.g., smart home, smart transportation, and autonomous vehicle systems.
Speaker: Yoshua Bengio, University of Montreal, Canada
Time: 10:00 am - 11:00 am Mar 19, 2018
Location: MakerSpace EventSpace, 6 MetroTech Center, Brooklyn, NY, US
Abstract: GANs and Unsupervized Representation Learning
One of the central questions for deep learning is how a learning agent could discover good representations in an unsupervised way. First, we consider the still open question of what constitutes a good representation, with the notion of disentangling the underlying factors of variation. Second, we discuss issues with the maximum likelihood framework which has been behind our early work on Boltzmann machines as well as our work on auto-regressive and recurrent neural networks as generative models. These issues motivated our initial development of Generative Adversarial Networks, a research area which has greatly expanded recently. We summarize recent GAN work aiming at better dealing with discrete data, as well as work aiming at generalizing GAN ideas to generative models which are iterative, like Boltzmann machines and denoising auto-encoders. We discuss how adversarial training can be used to obtain invariances to some factors in the representation, and a way to make training with such an adversarial objective more stable by pushing the discriminator score towards the classification boundary but not past it. Finally, we discuss applications of GAN ideas to estimate, minimize or maximize mutual information, entropy or independence between random variables.
About the Speaker: Yoshua Bengio (computer science, 1991, McGill U; post-docs at MIT and Bell Labs, computer science professor at U. Montréal since 1993): he authored three books, over 300 publications (h-index over 100, over 100,000 citations), mostly in deep learning, holds a Canada Research Chair in Statistical Learning Algorithms, is Officer of the Order of Canada, recipient of the Marie-Victorin Quebec Prize 2017, he is a CIFAR Senior Fellow and co-directs its Learning in Machines and Brains program. He is scientific director of the Montreal Institute for Learning Algorithms (MILA), currently the largest academic research group on deep learning. He is on the NIPS foundation board (previously program chair and general chair) and co-created the ICLR conference (specialized in deep learning). He pioneered deep learning and his goal is to uncover the principles giving rise to intelligence through learning, as well as contribute to the development of AI for the benefit of all.
Speaker: Sibin Mohan, University of Illinois at Urbana-Champaign
Time: 11:00 am - 12:00 pm Mar 29, 2018
Location: 2 MetroTech Center, 10th floor, Room 10.099, Brooklyn, NY
Abstract: Applications in the cyber-physical systems (CPS) domain are increasingly using commodity-off-the-shelf (COTS) components and are being interconnected for efficiency, better monitoring, and improved functionality. Traditionally, such systems were immune to software security attacks, but the increase in COTS components and interconnectivity are opening up new attack surfaces. Recent events, such as the Stuxnet incident, have shown the serious damage that can result from successful attacks. This can be particularly destructive for systems that have safety-critical constraints (e.g. avionics, automobiles, UAVs, power grids, etc.). In this talk, I intend to demonstrate how timing information can be crucial in the security of CPS especially when such systems have real-time properties.
In the first piece of work, I will demonstrate how timing information can be used to attack such systems – to leak critical information (e.g. the precise schedule and timing properties of tasks). In the second part of the talk, I will demonstrate multiple techniques that can deter attackers, both from the aforementioned attacks and also zero-day attacks that modify behavioral properties of such systems. I will also touch upon other timing-based defensive mechanisms that improve the overall security of safety-critical cyber-physical systems – both at design time, as well as retroactively (for legacy systems).
About the Speaker: Sibin Mohan is a Research Assistant Professor in both, the Dept. of Computer Science as well as the Information Trust Institute (ITI) at the University of Illinois. He completed his Ph.D. and M.S. in Computer Science from North Carolina State University in 2008 and 2004 respectively. His undergraduate degree was in Computer Science and Engineering from Bangalore University, India in 2001.
Sibin’s research interests are in the area of systems and security. His current research efforts include the integration of security in real-time embedded systems, intrusion detection in cyber-physical systems with real-time properties, secure cloud computing and the use of software defined networking (SDN) in safety-critical systems. In the past, he has done extensive work on the analysis of real-time and embedded systems and the development of system composition and safety techniques for avionics and medical devices.
He was previously a postdoctoral scholar in the Computer Science department at UIUC. In the past, he has also worked in Hewlett Packard’s India Software Operations.
ECE Seminar Series on Modern Artificial Intelligence: The Information Knot Tying Sensing and Action; Emergence Theory of Representation Learning
Speaker: Stefano Soatto
Time: 11:00 am - 12:00 pm Apr 5, 2018
Location: Pfizer Auditorium, 5 MetroTech Center, Brooklyn, NY, US
Abstract: Representations are functions of past data useful to accomplish future decision or control tasks. Ideally, they should be as informative as the data (sufficient), unaffected by nuisance factors in future data (invariant), as simple as possible (minimal), and easy to work with (disentangled). Such ideal representations are what one should store in memory in lieu of past data. But do they exist? If so, can they be computed? or learned? Minimality and sufficiency can be achieved by optimizing the Information Bottleneck Lagrangian, but how to do so? And what about invariance and disentanglement?
At face value, these classical principles from statistical decision and information theory have little to do with Deep Learning, where an empirical decision criterion is optimized with respect to a biologically-inspired (parametric) family of functions using stochastic gradient descent (SGD). Despite its simplicity, however, SGD has some surprising properties: First, it does not converge in the classical sense, but instead exhibits limit cycles that can be far from the critical points of the empirical loss. Second, it induces a bias - or regularization - in the learning process that is reminiscent of the Information Bottleneck Lagrangian, but not the usual one: This new one measures the information the parameters of the network (weights) contain about past data. The ideal properties we want, however, pertain to future data. What is the relation between these two information bottlenecks, past and future?
The Emergence Theory shows that minimizing the information the weights of a deep neural network contain about past data bounds minimality, invariance and disentanglement of the resulting representation of future data (activations). The resulting bound can be derived equivalently using Information Theory, or from PAC-Bayes theory. So, (explicit or implicit) regularization of the empirical loss used in Deep Learning provably induces the emergence of desirable properties of the representation implemented. I will discuss examples in visual recognition and control.
About the Speaker: Stefano Soatto is Professor of Computer Science and Electrical Engineering, and Director of the UCLA Vision Lab, in the Henry Samueli School of Engineering and Applied Sciences at UCLA. He is also Director of Applied Science at Amazon AI - AWS. He received his Ph.D. in Control and Dynamical Systems from the California Institute of Technology in 1996; he joined UCLA in 2000 after being Assistant and then Associate Professor of Electrical and Biomedical Engineering at Washington University, and Research Associate in Applied Sciences at Harvard University. Between 1995 and 1998 he was also Ricercatore in the Department of Mathematics and Computer Science at the University of Udine - Italy. He received his D.Ing. degree (highest honors) from the University of Padova- Italy in 1992. Dr. Soatto is the recipient of the David Marr Prize for work on Euclidean reconstruction and reprojection up to subgroups. He also received the Siemens Prize with the Outstanding Paper Award from the IEEE Computer Society for his work on optimal structure from motion. He received the National Science Foundation Career Award and the Okawa Foundation Grant. He was Associate Editor of the IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI) and a Member of the Editorial Board of the International Journal of Computer Vision (IJCV) and Foundations and Trends in Computer Graphics and Vision, Journal of Mathematical Imaging and Vision, SIAM Imaging. He is a Fellow of the IEEE.
Speaker: Vladimir Vapnik
Time: 10:00 am - 11:00 am May 4, 2018
Location: Pfizer Auditorium, 5 MetroTech Center, Brooklyn, NY, US
Abstract: The Seminar Series in Modern Artificial Intelligence begins a new tradition at New York University. The series will be held at NYU Tandon School of Engineering and is hosted by the Department of Electrical and Computer Engineering. Organized by Professor Anna Choromanska, the series aims to bring together faculty and students to discuss the most important research trends in the world of AI. The speakers include world-renowned experts whose research is making an immense impact on the development of new machine learning techniques and technologies and helping to build a better, smarter, more-connected world.
Speaker: Mehmet Toy, Verizon
Time: 11:00 am - 12:00 pm May 3, 2018
Location: 2 MetroTech Center, 10th floor, Room 10.099, Brooklyn, NY
Abstract: Cloud Services Architecture has been defined by Open Cloud Connect (OCC) standards organization which was merged with Metro Ethernet Forum (MEF) in 2016. Further work in this area has been initiated in ETSI Network Function Virtualization (NFV) and MEF.
In my talk, I will describe Cloud Services and their architecture, interfaces between a user and Cloud Service Provider, and interfaces between Cloud Service Providers. In addition, I will describe the high availability layers and fault management architecture of virtualized systems and services and relationships between failure recovery timers of the layers.
About the Speaker: Mehmet Toy holds Ph.D degree in Electrical and Computer Engineering from Stevens Institute of Technology, Hoboken, NJ, and M.S and B.S degrees in Electronics and Communications from Istanbul Technical University, Istanbul, Turkey.
He is a Distinguished Member of Technical Staff in Verizon Communications and involved in the implementation, testing and standards for SDN, NFV and Cloud Services.
Prior to this position, Dr. Toy has held senior technical and management positions in several well-known companies and startups including Comcast, Intel Corp., Verizon Wireless, Fujitsu Network Communications, AT&T Bell Labs and Lucent Technologies. He has also been a tenure-track faculty and adjunct professor in several universities including Stevens Institute of Technology, New Jersey Institute of Technology, Worchester Polytechnic Institute, and University of North Caroline at Charlotte.
Dr. Toy contributed to research, development and standardization of Cloud, Overlay, Self-Managed, SDN and Virtualized Commercial Networks and Services, Carrier Ethernet, IP Multimedia Systems (IMS), Optical, IP/MPLS, Wireless, ATM, and Signal Processing technologies. He holds a patent and has six pending patent applications. He has also published numerous articles, seven books and a video tutorial in these areas. Two of his books are being used as college text books and one of them is translated to Turkish.
Dr. Toy has served in the Open Cloud Connect Board, the IEEE Network Magazine Editorial Board, the IEEE Communications Magazine as a Guest Editor, the IEE-USA and the IEEE ComSoc in various capacities. He has received various awards from Comcast, AT&T Bell Labs and IEEE-USA for his accomplishments in these fields. He is currently a Sr. Member of IEEE and chair person of the IEEE ComSoc Cable Networks and Services sub-committee, and serves in MEF and ETSI NFV.
Stochastic-Robust and Robust Programs for the Ramp-Constrained Economic Dispatch Problem with Uncertain Renewable Energy
Speaker: Alberto J. Lamadrid, Lehigh University
Time: 11:00 am - 12:00 pm Apr 6, 2018
Location: 2 MetroTech Center, Room 9.011, Brooklyn, NY
Abstract: The inherent uncertainty of renewable energy sources (RES) makes the solution to the electricity network’s associated economical dispatch (ED) problem with network constraints challenging. In particular, the uncertainty in the power output of RES requires conventional generation units to ramp up and down more frequently to maintain the power balance and the reliability of the system. Typically, the RES power output uncertainty is modeled in ED problems by considering its potential future scenarios. However, this leads to an optimization problem that is difficult to solve for real-sized networks. Here, we present two proposals for this problem.
In the first one, we consider the uncertainty of RES and the consequent ramping of conventional generation via a robust reformulation of the problem. In particular, we show that in typical instances of the ED problem, the associated deterministic formulation of the robust problem can be solved efficiently for medium scale constrained electricity networks even when the underlying uncertainty distribution is not normal. Moreover, by comparing the proposed robust solutions to the ED problem with the typical scenario optimization approach, we show that the former solutions result on dispatch solutions that require less ramping than the later solutions, with little trade-off on the long-term expected costs of the dispatch. These results also provide insights about how RES penetration affects cost and dispatch policies in the electricity network. To illustrate our results, we present relevant numerical experiments on IEEE test networks.
In the second, We present an implementation of a two-stage security constrained unit commitment program with a recourse to dispatch the electricity system using a hybrid method to determine reserves for the System Operator (SO). This works is related to the stochastic optimization literature, with an emphasis on the economic determination and appraisal of different types of balancing reserves. The recourse decision balances the power dispatches, subject to the endogenously determined reserves. We study the implementation of the second settlement over a set of possible realized trajectories, as part of the implementation of a receding (rolling) horizon settlement.
About the Speaker: Alberto J. Lamadrid (Ph.D. Applied Economics and Management, Cornell University, 2012; M.A. Economics NYU, 2004; B.Sc. Electrical Engineering, Universidad de los Andes, Colombia, 2001) is an assistant professor with a joint appointment in the Economics Department and in the Industrial and Systems Engineering Department at the P.C. Rossin College of Engineering and Applied Science at Lehigh University. He is also a member of the Integrated Networks for Electricity Cluster at Lehigh. He has participated in NSF funded grants, as well as other awards funded by the Department of Energy, The Pennsylvania Infrastructure Technology Alliance and EPRI. His research interests are in electricity markets, power systems, and energy economics. He has worked on topics involving multi-period stochastic optimization in electrical networks, adoption of renewable energy sources and the valuation of power infrastructure assets.
Speaker: Nambi Seshadri, University of California, San Diego
Time: 11:00 am - 12:00 pm Apr 12, 2018
Location: MakerEvent Space in 6MTC, Rogers Hall
Abstract: Major strides have been made in the eld of wireless communications over the last century. In this work, I will present my personal take (and invite the audience to participate) on the key technical contributions that have impacted the practice of modern wireless communica- tions over the last ve decades. I am fortunate to have worked in this eld since my graduation and and will discuss my journey as well.
About the Speaker: Nambi Seshadri is currently a Professor in the School of Electrical Engineering at the University of California, San Diego, Chief Technology Of cer (Consulting) at Quantenna Communications and an advisor to several start ups in California and India. Prior to his current appointments, he was Chief Technology Of cer for Mobile and Wireless at Broadcom Corporation. From 1986-1999, Nambi was at AT&T Bell Labs and AT&T Shannon Labs where he conducted research on various aspects of wireless communications. He received his M.S. and Ph.D. from RPI in 1984 and 1986 respectively and his B.E. in Electronics and Communications from the Regional Engineering College, Trichy, India in 1982. He is a member of the US National Academy of Engineering and Foreign Member of Indian National Academy of Engineering. He was a co-recipient of the 1999 IEEE Information Theory Society Best Paper Award (with Vahid Tarokh and Rob Calderbank) for his work on Space-Time Trellis codes. He is a Fellow of IEEE and recipient of the 2018 IEEE Alexander Graham Bell Medal.
Speaker: Nicolo Michelusi, Purdue University
Time: 11:00 am - 12:00 pm Apr 16, 2018
Location: 2 MetroTech Center, Room 9.101, Brooklyn
Abstract: Mobile data traffic is expected to increase tremendously over the next decade, and cannot be accommodated by the limited bandwidth availability below 6GHz. On the other hand, the so called millimeter wave frequencies in the 28-100 GHz range promise to overcome these limitations. However, millimeter wave systems require narrow beam communication to achieve high throughput, thus mobile users need to be tightly tracked to provide seamless communication. Such requirement may pose severe challenges in mobile environments and may entail a significant performance degradation due to the associated signaling overhead. In the first part of the talk, we address the energy efficient design of the beam alignment protocol, with the goal of minimizing power consumption under communication constraints. We prove the optimality of a fractional search method, which senses a given fraction of the beam in each slot during beam alignment, and derive the fractional value in closed form. We also investigate the trade-off among beam-alignment, communication and user mobility. Beam alignment is achieved via a proper beam sensing protocol, which specifies how to allocate amplitude and phase at each antenna array element (a codeword) to sense the mobile user's position, through appropriate beam pointing. However, beam imperfections -- such as the presence of side-lobes -- and noise may cause errors in the detection process. Thus, in the second part of the talk, we investigate a Neyman-Pearson codebook design with optimal detection performance. We show that the optimal codebook is the principle eigenvector of a weighted array response matrix, and the dual problem can be solved via semidefinite programming. We show numerically that the proposed design outperforms a state-of-the art algorithm, with improvement up to 33% in detection performance.
About the Speaker: Dr. Nicolo Michelusi received the B.Sc. degree with honors, M.Sc. degree with honors and Ph.D. degree in Electrical Engineering from University of Padova, Italy, in 2006 and 2009, and 2013 respectively, and the M.Sc. degree in Telecommunication Engineering from Technical University of Denmark in 2009. In 2013-2015, he was a postdoctoral research fellow at the Ming-Hsieh Department of Electrical Engineering, University of Southern California, USA. He is currently an Assistant Professor at the School of Electrical and Computer Engineering, Purdue University, IN, USA, Associate Editor for the IEEE Transactions on Wireless Communications, and Senior Member of IEEE. His research interests lie in the areas of wireless communications, cognitive networks, energy harvesting IoT, 5G millimeter-wave networks, distributed algorithm over wireless networks, and machine learning applied to wireless communications. His research is currently funded by the National Science Foundation under grant CNS-1642982 and by DARPA to compete on the Spectrum Collaboration Challenge.
Speaker: Eby G. Friedman
Time: 11:00 am - 12:00 pm Apr 19, 2018
Location: MakerEvent Space in 6MTC, Rogers Hall
Abstract: The intention of this presentation is to provide an overview of the different projects of current focus in the high performance integrated circuit design research laboratory at the University of Rochester. Each of these topics considers different aspects of the systems integration process, with a focus on lower physical and circuit level aspects. Emphasis is placed on those fundamental challenges in delivering performance to high speed, high complexity heterogeneous integrated circuits. Technologies range from deeply scaled CMOS to emerging devices and circuits such as spintronic, photonic, and superconductive behaviors.
Delivering high quality power to on-chip circuitry with minimum energy loss is a fundamental objective of all modern integrated circuits (ICs). To supply sufficient power on-chip, an unregulated DC voltage is usually stepped down and regulated within the power delivery system. Power conversion and regulation resources should be efficiently managed to supply high quality power with minimum energy losses within multiple on-chip power domains. To satisfy challenging power efficiency and regulation requirements, hundreds of power regulators will be co-designed with many thousands of decoupling capacitors, distributing the power locally to billions of on-chip loads. Circuits, algorithms, and design methodologies are being developed to fundamentally change the manner in which power is delivered on-chip.
Three-dimensional (3-D) integration is changing the path for device scaling, supporting the delivery of multi-faceted heterogeneous systems. A variety of different design techniques and methodologies are under development to better design, model, architect, and build 3-D systems. Several test circuits have been developed to evaluate some of the key issues in 3-D system integration. These efforts will be reviewed and trends discussed.
Spintronic circuits have the potential to enhance CMOS in several dimensions, particularly as non-volatile memory and novel non-von Neumann structures. A variety of models and circuits will be described and placed within a CMOS perspective.
The energy expended in server farms has become an issue of seminal significance. CMOS simply expends too much energy to scale the size and number of these farms to support expected needs. An ultra-low energy technology is needed. One possible technology is superconductive single flux quantum (SFQ) circuits. This technology will be briefly reviewed, and novel design methodologies will be described to support the development of large scale Josephson junction based integrated systems.
About the Speaker: Eby G. Friedman received the B.S. degree from Lafayette College in 1979, and the M.S. and Ph.D. degrees from the University of California, Irvine, in 1981 and 1989, respectively, all in electrical engineering.
From 1979 to 1991, he was with Hughes Aircraft Company, rising to the position of manager of the Signal Processing Design and Test Department, responsible for the design and test of high performance digital and analog integrated circuits. He has been with the Department of Electrical and Computer Engineering at the University of Rochester since 1991, where he is a Distinguished Professor, and the Director of the High Performance VLSI/IC Design and Analysis Laboratory. He is also a Visiting Professor at the Technion - Israel Institute of Technology. His current research and teaching interests are in high performance synchronous digital and mixed-signal microelectronic design and analysis with application to high speed portable processors, low power wireless communications, and power efficient server farms.
He is the author of more than 500 papers and book chapters, 16 patents, and the author or editor of 18 books in the fields of high speed and low power CMOS design techniques, 3-D design methodologies, high speed interconnect, and the theory and application of synchronous clock and power distribution networks. Dr. Friedman is the Editor-in-Chief of the Microelectronics Journal, a Member of the editorial boards of the Journal of Low Power Electronics and Journal of Low Power Electronics and Applications, and a Member of the technical program committee of numerous conferences. He previously was the Editor-in-Chief and Chair of the Steering Committee of the IEEE Transactions on Very Large Scale Integration (VLSI) Systems, the Regional Editor of the Journal of Circuits, Systems and Computers, a Member of the editorial board of the Proceedings of the IEEE, IEEE Transactions on Circuits and Systems II: Analog and Digital Signal Processing, IEEE Journal on Emerging and Selected Topics in Circuits and Systems, Analog Integrated Circuits and Signal Processing, and Journal of Signal Processing Systems, a Member of the Circuits and Systems (CAS) Society Board of Governors, Program and Technical chair of several IEEE conferences, and a recipient of the IEEE Circuits and Systems Charles A. Desoer Technical Achievement Award, a University of Rochester Graduate Teaching Award, and a College of Engineering Teaching Excellence Award. Dr. Friedman is a Senior Fulbright Fellow and an IEEE Fellow.
Speaker: Gabriel Weaver, University of Illinois- Urbana-Champaign
Time: 1:30 pm - 2:30 pm Apr 19, 2018
Location: 2 MetroTech Center,Room 10.099, Brooklyn, NY
Speaker: Mei Chen, State University of New York
Time: 11:00 am - 12:00 pm Apr 20, 2018
Location: 2 MetroTech Center,Room 9.011, Brooklyn, NY
Speaker: Joerg Widmer
Time: 10:00 am - 11:00 am May 23, 2018
Location: 9.101 Conference Room, 9th floor, 2 MetroTech Center, Brooklyn
Abstract: The high bandwidth available millimeter-wave frequencies allows for very high data rates, and the latest wireless technologies, such as IEEE 802.11ad, are already starting to exploit this part of the radio spectrum to achieve rates of several GBit/s. Communication at these frequencies typically uses directional antennas which brings about interesting challenges to align antenna beams. Given the high penetration loss, most obstacles (e.g., a person) also completely block the signal. This results in a very dynamic radio environment and channels may appear and disappear over very short time intervals. This talk will give an overview of some of our recent works that deal with these networking challenges, focusing in particular on practical testbed research.
The talk covers low overhead continuous beam tracking on an 802.11ad compliant system with a phased array and an implementation of compressive beam training on commercial off-the-shelf devices. The very large bandwidth at mm-wave frequencies also allows, in principle, to design highly accurate location systems. Such location information can in turn be used to facilitate beam training, optimize access point association, predict (and counteract) future blockage, etc. However, implementing a location system based standard protocols and current mm-wave consumer hardware is challenging. The talk will present a location system we implemented on off-the-shelf 802.11ad devices, that only uses coarse angle information extracted from the beam training process. We further present a system (based on a custom hardware platform) that provides simultaneous localization and mapping of the environment, again based only on angle information.
About the Speaker: Joerg Widmer is Research Director as well as Research Professor at IMDEA Networks in Madrid, Spain. His research focuses on wireless networks, ranging from extremely high frequency millimeter-wave communication and MAC layer design to mobile network architectures. From 2005 to 2010, he was manager of the Ubiquitous Networking Research Group at DOCOMO Euro-Labs in Munich, Germany, leading several projects in the area of mobile and cellular networks. Before, he worked as post-doctoral researcher at EPFL, Switzerland on ultra-wide band communication and network coding. He was a visiting researcher at the International Computer Science Institute in Berkeley, USA, University College London, UK, and TU Darmstadt, Germany. Joerg Widmer authored more than 150 conference and journal papers and three IETF RFCs, and holds 13 patents. He serves or served on the editorial board of IEEE Transactions on Mobile Computing, IEEE Transactions on Communications, Elsevier Computer Networks and the program committees of several major conferences. He was awarded an ERC consolidator grant, the Friedrich Wilhelm Bessel Research Award of the Alexander von Humboldt Foundation, a Mercator Fellowship of the German Research Foundation, a Spanish Ramon y Cajal grant, as well as seven best paper awards. He is senior member of IEEE and ACM.