PhD Colloquia

Spring 2016

Seminars will be held in Towne 337 on Wednesdays at 12:00 PM unles otherwise specified.


January 13
Mohammad Amin Rahimian
Bayesian Learning without Recall
Click here to view Mohammad Amin Rahimian's seminar

Read the Abstract and Bio

Abstract: We analyze a model of learning and belief formation in networks in which agents follow Bayes rule yet they do not recall their history of past observations and cannot reason about how other agents' beliefs are formed. They do so by making rational inferences about their observations which include a sequence of independent and identically distributed private signals as well as the actions of their neighboring agents at each time. Successive applications of Bayes rule to the entire history of past observations lead to forebodingly complex inferences: due to lack of knowledge about the global network structure, and unavailability of private observations, as well as third party interactions preceding every decision. Such difficulties make Bayesian updating of beliefs an implausible mechanism for social learning. To address these complexities, we consider a Bayesian without Recall model of inference. On the one hand, this model provides a tractable framework for analyzing the behavior of rational agents in social networks. On the other hand, this model also provides a behavioral foundation for the variety of non-Bayesian update rules in the literature. We present the implications of various choices for the structure of the action space and utility functions for such agents and investigate the properties of learning, convergence, and consensus in special cases.

Bio: M. Amin Rahimian is a recipient of gold medal in 2004 Iran National Chemistry Olympiad. He was awarded an honorary admission to Sharif University of Technology, where he received his B.Sc. in Electrical Engineering-Control. In 2012, he received his M.A.Sc. from Concordia University in  Montréal, and he is currently a PhD student at the Department of Electrical and Systems Engineering and the GRASP Laboratory, University of Pennsylvania. His research interests include network science, distributed control and learning theory



January 20
Saad Aleem
Self-triggered Pursuit & Evasion
Click here to view Saad Aleem's seminar

Read the Abstract and Bio

Abstract: Pursuit and evasion strategies are widely observed in nature and play an important part in shaping predator-prey behaviors. In engineering, such problems have been the subject of much attention in combat games and more recently in the study of robotic systems for search and rescue missions and motion planning involving adversarial elements. Traditionally, treatment of this problem assumes continuous or periodic availability of sensing/communication on the part of the agents, which entails numerous unwanted drawbacks like increased energy expenditure in terms of sensing requirement, network congestion, inefficient bandwidth utilization, increased risk of exposure to adversarial detection, etc. In contrast, we are interested in the scenario where we can relax this continuous/periodic sensing requirement for the pursuer and replace it with triggered decision making, where the pursuer autonomously decides when it needs to sense the evader and update its trajectory to guarantee capture of the evader. In this talk, we study a pursuit-evasion problem involving a single pursuer and a single evader, where we are interested in developing a pursuit strategy that doesn’t require continuous, or even periodic, information about the position of the evader. We propose a self-triggered control strategy that allows the pursuer to sample the evader’s position autonomously, while satisfying desired performance metric of evader capture. To this end, we propose a self-triggered control strategy such that the pursuer can autonomously decide, based on out-dated information, when new samples of the evader’s position is required in order to satisfy desired performance metrics. We further extend this framework to develop a robust algorithm which allows for uncertainties in sampling the information about the evader, and derive tolerable upper-bounds on the error such that the pursuer can guarantee capture of the evader. In addition, we outline the advantages of retaining the evader’s history in improving the current estimate of the true location of the evader that can be used to capture the evader with even less samples. Our approach is in sharp contrast to the existing works in literature and our results ensure capture without sacrificing any performance in terms of guaranteed time-to-capture, as compared to classic algorithms that assume continuous availability of information.

 Saad Aleem received his BS in Electrical Engineering from the Lahore University of Management Sciences, Lahore, Pakistan, in 2013. He is currently pursuing the Ph.D. degree in Electrical and Systems Engineering at the University of Pennsylvania. His research interests are relevant to the area of multi-agent control with keen focus on triggered control, sampled-data control and fault diagnosis.



February 3
Brock Peterson
Laser Micromachined Magnets, High Fields and Field Gradients
Click here to view Brock Peterson's seminar

Read the Abstract and Bio

Abstract: High-energy-product rare-earth magnetic materials (such as SmCo and NdFeB alloys) have enabled or enhanced many application areas. In recent years, MEMS-scale applications have fueled interest in sub-millimeter microsctructured magnetics, where the overall magnet size or features on the magnet may range from micrometers to hundreds or micrometers. In addition, many MEMS systems can benefit from dense, alternating arrays of permanent magnets (PM) with high energy-product and substantial magnetic flux density adjacent to the array (applications include microgenerators, actuators, microundulators, etc.). We are developing such magnetic structures using a combination of laser micromachining and advanced assembly technologies. I will discuss the fabrication of these magnetic microstructures, some results from characterization and testing of the magnet arrays, and some of their possible applications.

Bio: Brock Peterson was born and raised north of Seattle, Washington. Brock received his B.S. in Electrical Engineering from Brigham Young University in August 2010. At BYU, he was involved in research studying sodium-based MRI designing RF resonant coils. Since then he had been pursuing his M.S. and Ph.D. degrees at Georgia Tech until transferring to Penn with Dr Allen at the beginning of 2014. Brock began his research with the MicroSensors and MicroActuators group in May 2011. The focus of his research has been on laser machined micromagnets and the applications of highly oscillating, strong magnetic field patterns.

February 10
Achin Jain
Optimal Control of a Hybrid Electric Vehicle with an Electrically Assisted Turbocharger

Read the Abstract and Bio

Abstract: In the wake of growing environmental concerns, the research focus in automobiles has shifted more towards electric or hybrid electric vehicles (HEV). We investigate the suitability of using an electrically assisted turbocharger in an HEV) with a turbocharged engine, based on fuel economy and acceleration performance. This system has two electric machines, a traction motor and a boost motor coupled to the shaft of the turbocharger, and offers an additional control variable in the energy management problem i.e. the amount of electrical boost (e-boost) to reduce the turbolag. The task of an optimal energy management controller now becomes manifold: deciding the torque split between the engine and the traction motor, the power of the boost motor and the gear number. We use dynamic programming to solve this optimal control problem and further analyze the circumstances under which it is advantageous to use the boost motor in an HEV.

Bio: Achin is a doctoral student in Electrical and Systems Engineering at the University of Pennsylvania. He received M.Sc. in Robotics, Systems and Control from ETH Zurich, Switzerland in 2015 and B.Tech. in Mechanical Engineering from the Indian Institute of Technology Delhi, India in 2012. His research interests include controls, optimization, statistics, machine learning applied to robotics, transportation and energy systems. His current research focuses on methods for predictive control with data-driven models.

February 17
Omur Arslan
Clustering-Based Robot Navigation and Control
Click here to view Omur Arslan's seminar

Read the Abstract and Bio

Abstract: With the increasing use of robots in our daily lives, it becomes even more crucial for autonomous robotics systems to be able to safely move in their workspaces to accomplish given tasks. Two widely used approaches to tackle the safe robot navigation problem are configuration space and sampling-based motion planning methods. While configuration space motion planning requires an explicit representation of the robot's configuration space (all possible allowed robot states that are free of any collisions, and satisfy the kinematic and dynamic system constraints) in terms of standard geometric and topological shapes, sampling-based motion planning avoids this complexity by producing open-loop navigation trajectories using randomly sampled robot configurations and simple connectivity criteria. In this talk, I will introduce the use of clustering for closing the gap in modeling configuration spaces between these two complementary motion planning methods. Traditionally an unsupervised learning method, clustering offers automated tools to discover coherent groups (clusters) in configuration spaces to model their unknown global organizational structure and/or to determine collision free neighborhoods of robot configurations. I will demonstrate potential applications of such clustering tools to the problem of feedback motion planning and control. In particular, I will present the use of hierarchical clustering for provably correct coordinated multirobot motion design, and I will show how the robot-centric Voronoi diagrams can be used for provably correct safe robot navigation in cluttered environments.

Biography: Omur Arslan received the B.Sc. and M.Sc. degrees in electrical and electronics engineering from the Middle East Technical University, Ankara, Turkey, in 2007 and from Bilkent University, Ankara, in 2009, respectively. He is currently working toward the Ph.D. degree with the Department of Electrical and Systems Engineering, University of Pennsylvania, Philadelphia. His current research interests include geometric and topological characterization of clustering methods and its application to robot motion planning and control, machine perception, machine learning and data mining.

February 24
Timothy Jones
Microwave Impedance Microscopy of Inhomogeneous and Nanostructured Materials
Click here to view Timothy Jones' seminar

Read the Abstract and Bio

Abstract: Atomic force microscopy and scanning tunneling microscopy were invented a few decades ago, but with modifications and added functionality have continued to provide a means of investigating nanoscale physical phenomena and the behavior of materials in response to electrical, magnetic, mechanical, chemical, and atomic forces. In microwave impedance microscopy (MIM), a mode of atomic force microscopy, a high frequency electrical signal is applied to the probe, focusing an evanescent wave at the tip apex, and the reflection is measured. The question of how electromagnetic fields interact with heterogeneous materials has been of historical interest, and with MIM we can revisit the question with a very high degree of localization in measurement. I use MIM, finite element simulations, and random resistor networks to test and model the response of porous and composite materials when exposed to focused, microwave electric fields and will discuss attempts to establish a correspondence between electric response and structural features such as porosity and pore size distribution.

Bio: In 2010, Timothy Jones received the B.S.E. in Electrical Engineering from the University of Pennsylvania. He has remained in the same department at Penn to study for the Ph.D. His current research involves the computational modeling of tip-sample interaction in functional atomic force microscopy techniques.


March 2 - 12:15 Start Time
Zhengwei Wu
Computationally Efficient Toeplitz-Constrained Blind Equalization Based on Independence
Click here to view Zhengwei Wu's seminar

Read the Abstract and Bio

Abstract: In blind equalization (BE), statistical or structural properties of the source symbols are used to recover the source from the observed convolution of the channel response and the source sequence, when no pilots are used and no information about the channel is available. In this work, we propose an algorithm based on independence with Toeplitz constraint (the T-EASI algorithm) at symbol rate to recover the source sequence. This scheme can give faster convergence than common BE methods. In addition, we improve upon the T-EASI algorithm which greatly decreases the computational cost. The T-EASI iterations can be efficiently implemented using the fast Fourier transform (FFT). At the same time, an approximation of the cross-correlation terms used in the adaptation also helps reduce computational complexity. For sources with independent I/Q components, the I/Q independence constraint can be used to reduce the phase ambiguity.

Bio: Zhengwei Wu is currently working towards the Ph.D. degree in the Department of Electrical and Systems Engineering at the University of Pennsylvania. Her advisor is Saleem A. Kassam. Zhengwei received the B.Sc. degree in electrical engineering from Shanghai Jiao Tong University, Shanghai, China, in 2011, and the M.Sc. degree in electrical engineering from the University of Pennsylvania, Philadelphia, in 2013. Her research interests include statistical signal processing, blind source separation, channel equalization, wireless communication and communication systems.

March 16
Yuan Li
Thick Multilayered Permanent Micromagnets with Preserved Magnetic Energy Density
Click here to view Yuan Li's seminar

Read the Abstract and Bio

Abstract: This talk presents microfabricated permanent magnets possessing a multilayer structure that preserves the high magnetic energy density of thinner magnetic films while simultaneously reducing average residual stress of the films and achieving a significant magnetic thickness. Many magnetic MEMS devices heavily rely on the availability of thick (a few tens to hundreds of micrometers), high-energy-density permanent magnet components able to be deposited in a fully-integrated and CMOS-compatible manner (process temperature less than 450 ?C). However, it is observed that increasing magnetic film thickness typically causes 1) a concomitant decrease in magnetic properties such as maximum energy density; and 2) increased mechanical instability (cracking and delaminating, due to the increased elastic strain energy stored in the films as magnetic volume increases), both of which limit the maximum total magnetic energy of these small-scale integrated magnets. The microlaminated permanent magnet presented here utilizes sequential multilayer electroplating, in which alternating layers of relatively thin magnetic films (CoNiP, micrometer range) and non-magnetic materials (Cu, a few tens of nanometer to micrometer range) are electrodeposited in a multilayer fashion realizing thick laminated permanent micromagnets with improved total magnetic energy as compared with their non-laminated counterparts. Low interface roughness has been demonstrated to play an important role on preserving the component magnetic thin layer properties in the multilayer configuration.

Bio: Yuan Li received his Mechanical Engineering from Tsinghua University, Beijing, China in 2009. He received his in Mechanical Engineering in Georgia Tech in 2012 with thesis focusing on mechanical behavior of nanostructured materials. Yuan is currently a PhD student in Electrical and System Engineering at Penn. His current research interest focuses on exploration of micro-magnet fabrication method for MEMS applications.

March 23
Rasul Tutunov
Distributed SDDM Solvers
Click here to view Rasul Tutunov's seminar

Read the Abstract and Bio

Abstract: We propose a distributed second order method for general consensus. Our approach is the fastest in literature so-far as it outperforms state-of-the-art methods, including ADMM, by significant margins. This is achieved by exploiting the sparsity pattern of the dual Hessian and transforming the problem to a one of efficiently solving a sequence of symmetric diagonally dominant system of equations. We validate the above claim both theoretically and empirically. On the theory side, we prove that similar to exact Newton, our algorithm exhibits superlinear convergence within a neighborhood of the optimal solution. Empirically, we demonstrate the superiority of this new method on a variety of machine learning problems and show that our improvements arrive at low communication overhead between processors.

Bio: Rasul Tutunov  received his B.Sc and M.A.Sc. in Physics and Applied Mathematics in 2007 and 2009 from Moscow Institute Of Physics and Technology. Currently he is a 5 year PhD student at the Computer and Information Science Department of University of Pennsylvania. His research interests include network science, distributed optimization, deep learning and spectral graph theory.

March 30 - 12:15 Start Time
Heejin Jeong
Learning Complex Stand-up Motion for Humanoid Robots
Click here to view Heejin Jeong's seminar

Read the Abstract and Bio

Abstract: There are a number of small-size humanoid robots with the ability to stand up after they fall down. However, most of their stand-up motions consist of several motion sequences in a fixed order with specific sets of joint angles. In these cases, since a robot always has to move to the desired pose of the first motion sequence regardless of its initial pose after falling, such stand up motions can lead to time delays in standing or cause damage to the robots from hitting the ground or receiving high torque forces. Adult-size humanoid robots can become even more seriously damaged by these motions in general. During the DARPA Robotics Challenge Final last year, many robots fell down multiple times, but only one robot managed to stand up after falling down. This shows that the ability to stand up from various poses in different environments becomes increasingly important to robots as they start to work in human environments or alongside humans. In this talk, I will introduce stand-up motions of a humanoid robot found by reinforcement learning. As an initial approach, we first considered an obstacle-free environment with flat and even ground, and we used a Darwin-OP humanoid robot for an application platform. We discretized continuous state and action spaces using Gaussian Mixture Model, and we used Q-learning for its learning method. An optimal policy learned in simulation has been applied to the real robot and resulted motions showed better performance than previously hand-designed stand-up motions. I will also talk about challenges in (humanoid) robot reinforcement learning that we have faced and the scalability of this approach to more complicated environment.

Bio: Heejin Jeong is a doctoral student in the Department of Electrical and Systems Engineering at the University of Pennsylvania. Her research interests include decision making under uncertainty, reinforcement learning, cognitive psychology, computational neuroscience and human-robot interaction in space exploration. Her current work focuses on designing decision making algorithm using cognitive models and applying reinforcement learning in humanoid robotics. She participated last summer in the DARPA Robotics Challenge Final as well as the 2015 Robocup league for adult-size humanoid robots. Her team won first place in 2015 Robocup competition. Before coming to Penn, she received her B.S. degree in Physics and Aerospace Engineering (minor) from Korea Advanced Institute of Science and Technology (KAIST) in 2014.

April 6 - 12:15 Start Time
Jeffrey Duperret
A Framework for Agile Legged Transitional Mobility

Read the Abstract and Bio

Abstract: The past five years have seen impressive advances in legged robots performing running behaviors, which in part is due to a strong theoretical understanding of stable steady-state gaits in reduced order dynamical systems.  On the other hand, controlling for non steady-state behaviors remain poorly understood -- particularly for agile behaviors requiring rapid changes in energy or direction such as accelerating, leaping, and turning.  Achieving such agile behaviors is one of the intuitive motivations for using legs in robotic locomotion and would greatly enhance the locomotion prowess of legged machines operating in highly unstructured environments such as in aiding disaster response.  This talk discusses recent work in controlling simple non steady behaviors leveraging the natural hybrid dynamics of the system as well as design choices regarding robot morphology (particularly in adding core actuation to the robot) that can aid in achieving energetic maneuvers.

Bio: Jeffrey Duperret is currently a doctoral student under Daniel Koditschek in the Department of Electrical and Systems Engineering at the University of Pennsylvania.  He received Bachelor of Science in Engineering degrees in electrical engineering and computer science from the University of Michigan in 2011, where he was involved in research on control techniques for autonomous car collision avoidance.  Jeff's current research interests are in the control of robotic legged locomotion, specifically investigating the utility of core actuation/compliance for quadrupedal platforms as well as developing formalisms for the description and control of non-steady-state agile behaviors.

April 13
David Sun
The Rise and Fall of Great Powers: Understanding Coalition Dynamics in a Competitive Setting with Extensive-Form Game and Agent-based Modeling Approach
Read the Abstract and Bio

Abstract: Through this research, we look at interacting agents with distinct attributes in multiparty competitive games and explore how the coalition dynamics would evolve from any certain initial state, through the development of a tractable framework and an agent-based model. In the broader context of multiparty competitive games (which are closely related to civil wars), we make several critical assumptions regarding the motivations for coalition formation (and dis-integration) and view coalitions as an outcome of rational, utilitarian choices made by the individual players. Accordingly, we investigate two specific types of such problems. The first problem is the existence and characteristics of the stable states. We investigate when such stable states are viable, and potential path dependency on initial states. The second problem is the influence of certain changes in agent properties (objective position in Euclidean space, relative power size, etc.) on the dynamics of coalition formations. We study both its influence on the set of realized stable states, as in how would such difference affects the eventual outcome, and its influence on the shift of stable states, as in how would such difference may lead to a potential state shift (when perturbed).

We approach the problems using two set of methods. First, we present a formal, mathematical model inspired by theories from non-cooperative game in its extensive form with varying settings, to prove the existence of and characterize the stable states. Later, we develop an agent-based counterpart built upon the former and use computational modeling to explore the system dynamics for the baseline model. The findings on certain patterns of coalition dynamics will be further tested with statistical methods and case studies under varying themes, and compared with a set of prominent theories from multiple fields.

Bio: Qiwei (David) Sun received his B.Sc. in Electrical Engineering and Computer Science from International University Bremen, Germany in 2012. He is currently pursuing the PhD degree in Electrical and Systems Engineering at the University of Pennsylvania, under the supervision of Prof. Barry Silverman at Ackoff Collaboratory for Advancement of the Systems Approach (ACASA) lab. His research interests include network science, the agent-based modeling & simulations, and decision theory.

April 20
Nick Watkins
Optimal Resource Allocation for Competitive Spreading Processes

Read the Abstract and Bio

Abstract: Understanding how beliefs and behaviors spread through networked populations of agents is important to the continued advancement of our understanding of societal behaviors. In this talk, we consider a formal mathematical model for the spread of competitive behaviors in a network, and develop a framework to find an optimal allocation of resources to effectively control a chosen behavior. We consider the SI1SI2S model, a recent generalization of the popular SIS model to the case of two competitive epidemics. We start our analysis by extending the standard SI1SI2S formulation with homogeneous parameters to a heterogeneous setting with edge dependent infection rates, and node-dependent recovery rates. We then find necessary and sufficient conditions under which the mean-field approximation of a chosen epidemic process stabilizes to extinction exponentially quickly. Leveraging this result, we develop a framework for the solution of two optimization problems. In the first, we find an optimal allocation of control resources in order to eradicate the chosen epidemic at a minimum cost. In the second, we are given a fixed budget and propose a method which provably attains the extinction condition when sufficient capital is available, and otherwise mitigates the spread of the unwanted epidemic as much as possible. We close the talk by investigating the accuracy of the assumptions made by our developments, and exploring possible mechanisms for making further progress.

Bio: Nicholas J. Watkins is currently a Ph.D. student in the Electrical and Systems Engineering program at the University of Pennsylvania.  He graduated summa cum laude with a B.S. in Electrical Engineering from Wilkes University in 2013.  His current research interests include the analysis and control of epidemic spreading processes, network science, cyber-physical systems and optimal control.

April 27
Cassiano Becker
Spectral Inference of Resting Functional Connectivity in Brain Networks

Read the Abstract and Bio

Abstract: Understanding the relationship between the dynamics of neural processes and the anatomical substrate of the brain is a central question in neuroscience. On the one hand, modern neuroimaging technologies, such as diffusion tensor imaging, can be used to construct structural graphs representing the architecture of white matter streamlines crisscrossing the cortex. On the other hand, temporal patterns of neural activity can be used to construct functional graphs representing temporal correlations between brain regions. Although some studies provide evidence that whole-brain functional connectivity is shaped by underlying anatomy, the observed relationship between function and structure is weak, and the rules by which anatomy constrains brain dynamics remain elusive. In this talk, we introduce a methodology to predict with high accuracy the functional connectivity of a subject at rest from his or her structural graph. Using our methodology, we are able to systematically unveil the role of indirect structural paths in the formation of functional correlations. Furthermore, in our empirical evaluations, we observe that the eigen-modes of the predicted functional connectivity are aligned with activity patterns associated to different cognitive systems.

Bio: Cassiano Becker is a PhD student in the Department of Electrical and Systems Engineering at the University of Pennsylvania. He earned an M.Sc. in Telecommunications (with distinction) from the University College London, and a B.Eng. in Electrical Engineering from the Federal University of Rio Grande do Sul, Brazil. Before starting his doctoral studies, he worked as systems engineer, software developer and project manager in technology companies such as Harris Corporation and Siemens. His research interests include applications of network models, dynamical systems and machine learning to neuroscience, brain-computer interfaces and other cyber-physical systems.

May 4
Vasileios Tzoumas
Sensor Placement for Optimal Kalman Filtering: Fundamental Limits, Submodularity, and Algorithms
Read the Abstract and Bio

Abstract: In this paper, we focus on sensor placement in linear dynamic estimation, where the objective is to place a small number of sensors in a system of interdependent states so to design an estimator with a desired estimation performance. In particular, we consider a linear time-variant system that is corrupted with process and measurement noise, and study how the selection of its sensors affects the estimation error of the corresponding Kalman filter. Our contributions are threefold: First, we prove among other design limits
that the number of sensors grows linearly with the system’s size for fixed minimum mean square estimation error and number of output measurements over an observation interval. Second, we prove that the logdet of the error covariance of the Kalman filter with respect to the system’s initial condition and process noise is a supermodular and non-increasing set function in the choice of the sensor set. Third, we provide efficient approximation algorithms that select a small number sensors so to design an optimal Kalman filter with respect to this estimation error —the worst-case performance guarantees of these algorithms are provided as well. Finally, we illustrate the efficiency of our algorithms using the problem of surface-based monitoring of CO2 sequestration sites studied in Weimer et al. (2008).

Bio: Vasileios Tzoumas received a degree in Electrical and Computer Engineering from the National Technical University of Athens, Greece, in 2012, where his Diploma thesis focused on competitive information diffusion processes over social networks from a game-theoretic perspective.  Currently, he is working toward the Ph.D. degree in the Department of Electrical and Systems Engineering, University of Pennsylvania, Philadelphia. His research interests include large-scale networked control systems, and the minimal actuator and sensor placement in them for optimal control and estimation.  He received the best session presentation award at the 2015 American Control Conference (ACC).


May 11
Greg Henselman
Topological Network Analysis: Theory & Applications

Read the Abstract and Bio

Abstract: The study of continuous maps between metric spaces plays an important role in most rigorous introductions to calculus and calculus-based modeling.  The larger field to which this subject belongs, algebraic topology, is often characterized as the study of abstract space “up to continuous deformation,” and plays a fundamental role in the modern and 20th century treatment of mathematical analysis, dynamics, mathematical economics, and the physical sciences.   Commensurate with the advance of computing resources in the late 20th century, a growing body of work has been devoted to the adaptation of topological methods for problems in discrete modeling, particularly in the context of coarse, relational, low-grade, and high-dimensional data, where invariance under deformation translates to robustness under noise, bias, and measurement error.  In this expository talk, I will discuss recent applications in random matrix theory, network coding, and hypothesis-testing, with brief historical remarks on the role of algebraic topology in complex analysis, maxwell’s equations, fMRI data , and the study of nematic liquid-crystals.  Time permitting, I will discuss recent advances in computation and discrete dynamics.

Biography: Gregory Henselman earned bachelor’s degrees in Mathematics and Classical Studies at Willamette University in 2010, where he was involved in the study of eigenvalue statistics of fixed-genus one-face maps, and a master’s degree in Mathematics from the University of Oregon in 2011, where he studied topological complexity theory with Professor Sergey Yuzvinsky.  He joined the PhD program at the University of Pennsylvania in 2011 under the direction of Professor Robert Ghrist, and his interests include cellular sheaf theory, computational topology, and discrete optimization.  His recent work applies cellular matroid theory to problems in shape description and homological persistence


Fall 2015

August 26
Konstantinos Gatsis
Opportunistic Resource Allocation for Wireless Control Systems
Click to view video of Konstantinos Gatsis' seminar

Read the Abstract and Bio

Abstract: This work is motivated by modern cyber-physical environments
appearing in building automation, industrial applications, and the smart
grid. These systems are characterized by multiple sensor and actuator
devices at different physical locations, communicating wirelessly with
each other. Desired closed loop performance requires efficient wireless
communication. However the latter is constrained by transmission
uncertainties, e.g., packet drops, as well as by limited resources
available to the devices, in particular transmit power. This work
focuses on the design of resource allocation policies that adapt to the
control requirements of the physical plant, and opportunistically
exploit the random wireless medium variations. Two important
developments are power management for single loop systems, as well
scheduling and power management for control systems over shared
(multiple access) wireless channels.

Bio:Konstantinos Gatsis received the Diploma degree in electrical and
computer engineering from the University of Patras, Patras, Greece in
2010. Currently, he is working toward the Ph.D. degree in the Department
of Electrical and Systems Engineering, University of Pennsylvania,
Philadelphia. His research interests include cyber-physical systems,
networked control systems, as well as resource optimization problems
arising in them. Mr. Gatsis received the 2014 O. Hugo Schuck Best Paper
Award, the Student Best Paper Award at the 2013 American Control
Conference, and was a Best Paper Award Finalist at the 2014 ACM/IEEE
International Conference on Cyber-Physical Systems (ICCPS).


September 2
Alec Koppel
Online Learning and Pattern Recognition in Multi-Agent Systems
Click to view video of Alec Koppel's seminar
Read the Abstract and Bio

An algorithm to learn optimal actions in convex distributed online problems is developed. Learning is online because cost functions are revealed sequentially and distributed because they are revealed to agents of a network that can exchange information with neighboring nodes only. Learning is measured in terms of the global network regret, which is defined here as the accumulated loss of causal prediction with respect to a centralized clairvoyant agent to which the information of all times and agents is revealed at the initial time. A variant of the Arrow-Hurwicz saddle point algorithm is proposed to control the growth of global network regret. This algorithm uses Lagrange multipliers to penalize the discrepancies between agents and leads to an implementation that relies on local operations and exchange of variables between neighbors. We show that decisions made with this saddle point algorithm lead to regret whose order is not larger than O(sqrt(T)), where T is the total operating time. Numerical behavior is illustrated for the particular case of distributed recursive least squares. An application to computer network security in which service providers cooperate to detect the signature of malicious users is developed to illustrate the practical value of the proposed algorithm.

We then consider an extension of this problem class to the case where agents aim to learn a common discriminative signal representation for a particular learning task, which makes the program non-convex. Nonetheless, we are able to establish the convergence in expectation of a block-stochastic saddle point algorithm for this setting. We apply the proposed method to the problem of a mobile robotic team seeking to collaboratively incorporate visual information into navigability assessment in an unknown environment, demonstrating the proposed algorithm's utility in a field setting.

Bio: Alec Koppel is a doctoral student in the Department of Electrical and Systems Engineering at the University of Pennsylvania and a participant in the Science, Mathematics, and Research for Transformation (SMART) Scholarship Program sponsored by the American Society of Engineering Education. His sponsoring facility is the U.S. Army Research Laboratory in Adelphi, MD, where he works during doctoral summers. Before coming to Penn, completed his M.S. degree in Systems Science and Mathematics and B.A. degree in Mathematics at Washington University in St. Louis, MO. His research interests are in the areas of signal processing, optimization, and learning theory. His current work focuses on designing new large-scale or dynamic optimization methods for multi-agent systems with a focus on robotic and computer networks.


September 9
Fei Miao
Data-Driven Robust Taxi Dispatch Approaches
Click to view video of Fei Miao's seminar
Read the Abstract and Bio

Abstract: Traditional transportation systems in metropolitan areas often suffer from inefficiencies due to uncoordinated actions as system capacity and traffic demand change. With the pervasive deployment of networked sensors in modern vehicles, large amounts of information regarding traffic demand and system status can be collected in real-time. This information provides opportunities to perform various types of control and coordination for large scale intelligent transportation systems. In this work, we first present a receding horizon control (RHC) framework to dispatch taxis, which combines highly spatiotemporally correlated demand/supply models and real-time GPS location and occupancy information.  The objectives include reducing taxi idle driving distance and matching spatiotemporal ratio between demand and supply for service quality. Such efficient dispatch control and coordinating strategies face a new challenge: how to deal with future customer demand uncertainties while fulfilling system's  performance requirements. To address this problem, we then present a novel robust optimization method for taxis dispatch problems to consider closed convex form of spatiotemporally correlated demand model uncertainties. The robust optimization problem is proved equivalent to a convex optimization form by strong duality and minimax theorem,  given uncertain demand sets, and computational tractability is guaranteed.  Extensive trace driven analysis with real taxi operational record data sets show that the RHC framework reduce the average total idle distance and reduce the total supply demand ratio error across the city. Moreover, the robust taxi dispatch solutions are less probable to get large costs compared with non-robust results.

Bio: Fei Miao received the B.Sc. degree in Automation from Shanghai Jiaotong University (SJTU), Shanghai, China, in 2010 and the M.A. degree in statistics from the dual degree program of Wharton, University of Pennsylvania, in 2015. Currently, she is working toward the Ph.D. degree in the Department of Electrical and Systems Engineering at University of Pennsylvania. Her research interests focus on the control aspect of Cyber-Physical Systems (CPS), include data-driven control frameworks of large-scale interconnected CPSs under model uncertainties, and resilient control frameworks to address security issues of CPSs. Ms. Miao was a Best Paper Award Finalist at the 2015 ACM/IEEE International Conference on Cyber-Physical Systems (ICCPS), CPSWeek



September 16
Santiago Paternain
Online Learning of Feasible Strategies in Unknown Environments
Click to view video of Santiago Paternain's seminar
Read the Abstract and Bio

An environment is defined as a set of constraint functions that vary arbitrarily over time. An agent wants to select feasible actions that keep all the constraints negative, but must do so causally. I.e., the dynamical system that determines actions is such that only their time derivatives can depend on the current constraints. An environment is said viable if there exists an action that can satisfy the constraints for all times. The fit of a trajectory is defined as a vector that integrates the constraint violations over time and is used to measure the extent to which a policy succeeds in learning feasible actions. An online saddle point controller is proposed to control fit and shown to do so under minimal technical conditions. The online saddle point controller pushes actions along a linear combination of the constraint negative gradients and dynamically adapts the coefficients of this linear combination to find appropriate weightings. Concepts are illustrated throughout with the problem of a shepherd that wants to stay close to all sheep in a herd. Numerical experiments show that the controller allows the shepherd to do so.

Bio:Santiago Paternain received the B.Sc. degree in electrical engineering from Universidad de la Republica Oriental del Uruguay, Montevideo, Uruguay in 2012. Since August 2013, he has been working toward the Ph.D. degree in the Department of Electrical and Systems Engineering, University of Pennsylvania. 


September 23
Towne 337
Arnob Ghosh
Economics of Secondary Spectrum Market
Click to view video of Arnob Ghosh's seminar
Read the Abstract and Bio

Abstract: The demand for wireless spectrum has increased because of the growth of smart phones and smart devices. Studies show that the demand will soon surpass the available spectrum. However, studies find that the licensed bandwidth is often used inefficiently. Hence, the research community is looking for the possibility where the license holders (primaries) can allow the unlicensed users (secondaries) to access the spectrum (known as secondary spectrum access). We investigate a secondary spectrum oligopoly market where multiple primaries will lease their channels to the secondaries in lieu of financial remuneration.

The secondary spectrum market provides significant challenges compared to the traditional markets because of the interference and the uncertainty of competition. Signals can not be transmitted simultaneously at the adjacent locations because of the interference. The quality of service provided by the primaries may vary and a primary is not aware of the quality of service provided by its competitors (known as uncertainty of competition). We model the above market as a non cooperative game with primaries as players. Each primary owns a channel over a network and it must select a set of non interfering locations and the price at each location depending on the quality of the channel. The Nash equilibrium that we have characterized has interesting properties and depend on the topology of the graph. 

Bio:Arnob Ghosh is currently a PhD student in the Electrical and Systems Engineering department. His advisor is Saswati Sarkar. Arnob obtained his M.S. in EE from UPENN and his Bachelor of Engineering in ECE from Jadavpur University in 2013 and 2011 respectively. His research interests includes Economics of spectrum sharing in Cognitive Radio Network, Resource allocation in Wireless network using Game theoretic approaches, and network economics.


September 30
Fragkiskos Koufogiannis
Gradual Release of Sensitive Data Under Differential Privacy
Click to view video of Fragkiskos Koufogiannis' seminar

Read the Abstract and Bio

We introduce the problem of releasing sensitive data under differential privacy when the
privacy level is subject to change over time. Existing work assumes that privacy level, denoted by epsilon, is determined by the system designer as a fixed value before any sensitive data is released. For certain applications, however, users may wish to relax the privacy level for subsequent releases of the same data after either a re-evaluation of the privacy concerns or the need for better accuracy. Specifically, given a database containing sensitive data, we assume that a response y1 that preserves epsilon1-differential privacy has already been published. Then, the privacy level is relaxed to epislon2 and we wish to publish a more accurate response y2 while the joint response (y1; y2) preserves epsilon 2-differential privacy. How much accuracy is lost in the scenario of gradually releasing two responses y1 and y2 compared to the scenario of releasing a single response that is epsilon2-differentially private? Our results show that there exists a composite mechanism that achieves no loss in accuracy.

We consider real-valued sensitive data and we, initially, focus on mechanisms that approximate the identity query. We derive a composite mechanism based on a lazy Markov stochastic process that performs gradual release of sensitive data without accuracy loss. Closed-form expressions and efficient sampling algorithms are provided for this stochastic process. The applicability of our results is demonstrated on several scenarios, including Google's project RAPPOR, trading of sensitive data, and controlled diffusion of private data in a social network.

Fragkiskos Koufogiannis is currently working towards his Ph.D. degree in the Department of Electrical and Systems Engineering, University of Pennsylvania, Philadelphia. In 2011, he received his Diploma in Electrical and Computer Engineering from Aristotle University of Thessaloniki, Thessaloniki, Greece, while he completed his diploma's thesis in Ecole Polytechnique Federale de Lausanne, Lausanne, Switzerland. His research interests include optimization and probability theory. His current work focuses on techniques for processing sensitive data. He works both on theoretical aspects of differential privacy by deriving optimality results, and on applied aspects by presenting privacy-aware algorithms.


October 7
Aryan Mokhtari
Stochastic Quasi-Newton Methods
Click to view video of Aryan Mokharti's seminar
Read the Abstract and Bio


The solution of stochastic optimization problems with stochastic gradient descent algorithms (SGD) is widespread, but SGD methods are slow to converge. This has motivated the use of stochastic quasi-Newton methods that utilize stochastic gradients as both, descent directions and ingredients of a curvature estimation methodology. We consider two methods: (i) RES, a regularized stochastic version of the BFGS method (ii) oLBFGS a stochastic limited memory version of BFGS. We show that both of these methods converge to optimal arguments under hypotheses of strong convexity and decreasing stepsizes. We further establish O(1/t) convergence rates in expectation and present numerical evaluations to showcase the advantages relative to SGD.


Aryan Mokhtari received the B. Eng. degree in electrical engineering from Sharif University of Technology, Tehran, Iran, in 2011, and the M.S. degree in electrical engineering from the University of Pennsylvania, Philadelphia, PA, in 2014. Since 2012, he has been working towards the Ph.D. degree in the Department of Electrical and Systems Engineering, University of Pennsylvania, Philadelphia, PA. His research interests include stochastic optimization, machine learning and distributed optimization.


October 14
Towne 337
Weiyu Huang
Metrics in the Space of High Order Networks
Click to view video of Weiyu Huang's seminar
Read the Abstract and Bio

We consider high order networks, defined as weighted complete hypergraphs collecting relationship functions between elements of tuples. They can be considered as generalizations of conventional networks where only relationship functions between pairs are defined. Two families of distances are introduced in the space of high order networks. The distances measure differences between networks. Such distances are difficult to compute when the number of nodes is large. To solve this problem, we relate high order networks to the filtrations of simplicial complexes and show that the dissimilarity between networks can be lower bounded by the difference between the homological features of their respective filtrations. Practical implications are explored by comparing the coauthorship networks of different research communities or different authors within the same community. The metrics and the lower bounds succeed in identifying the respective collaboration patterns of different authors and different research communities.

Bio: Weiyu Huang received the B.Eng. (Hons.) degree in electronics and telecommunication from the Australian National University (ANU), Canberra, Australia, in 2012. He is currently pursuing the Ph.D. degree in electrical and systems engineering at the University of Pennsylvania, Philadelphia, PA, USA. From 2011 to 2013, he was a Telecommunication Engineer and Policy Officer with the Australian Communication and Media Authority. His research interests include signal processing, network theory, pattern recognition, and the study of networked data arising in human, social, and technological networks. Mr. Huang was the recipient of the University Medal offered by the ANU.



October 21
Towne 337
Santiago Segarra
Diffusion Dynamics and Distributed Networking: A Graph Signal Processing Perspective
Click to view video of Santiago Segarra's seminar
Read the Abstract and Bio


A network can be understood as a complex system formed by multiple nodes, where global network behavior arises from local interactions between connected nodes. Often, networks have intrinsic value and are themselves the object of study. In other occasions, the network defines an underlying notion of proximity or dependence, but the object of interest is a signal defined on top of the graph. This is the matter addressed in the field of graph signal processing (GSP). Graph-supported signals appear in many engineering and science fields such as gene expression patterns defined on top of gene networks and the spread of epidemics over social networks. Transversal to the particular application, the philosophy behind GSP is to advance the understanding of network data by redesigning traditional tools originally conceived to study signals defined on regular domains and extend them to analyze signals on the more complex graph domain. In this talk, we will introduce the main building blocks of GSP and illustrate the utility of these concepts through real-world applications. Our focus will be on the definition and application of graph sampling and blind graph filter identification.

Santiago Segarra received the B.Sc. degree in industrial engineering with highest honors from the Instituto Tecnológico de Buenos Aires (ITBA), Argentina, in 2011 and the M.Sc. degree in electrical engineering from the University of Pennsylvania, Philadelphia, in 2014. Since 2011, he has been working towards the Ph.D. degree in the Department of Electrical and Systems Engineering at the University of Pennsylvania. His research interests include network theory, data analysis, machine learning, and graph signal processing. Mr. Segarra received ITBA’s 2011 award to the best undergraduate thesis in industrial engineering and the 2011 outstanding graduate award granted by the National Academy of Engineering of Argentina.

October 28
Towne 337
Mahyar Fazlyab
Optimal Design for Synchronization of Oscillator Networks
Click to view video of Mahyar Fazlyab's seminar
Read the Abstract and Bio


In this paper, we consider the problem of designing a network of nonidentical Kuramoto oscillators in order to achieve optimal synchronization in steady-state. We choose phase cohesiveness as the synchronization metric, defined as the maximum asymptotic phase difference across all the edges. We address two network design problems: (1) the nodal-frequency design problem, in which we design the natural frequencies to optimize phase cohesiveness for a fixed network structure, and (2) the robust edge-weight design problem, in which we design the link weights assuming that the natural frequencies belong to a given convex uncertainty set. In both problems, we associate a tuning cost function to the optimization variables. In this context, we develop a convex optimization framework to solve both the nodal frequency and the edge-weight design problems under budget constraints. We illustrate the applicability of the proposed framework by analyzing particular network synchronization problems, such as sparsity-promoting network design, robust network design for distributed wireless analog clocks, power re-dispatch in power grids, and the Braess’ paradox.

Mahyar Fazlyab is a PhD student in the ESE department, University of Pennsylvania. Before joining Penn in 2013, he got his Bachelors and Masters degree, both in Mechanical Engineering, from Sharif University of Technology, Tehran, Iran. His research interests include but not limited to the analysis, optimization, and control of (networked) dynamical systems.

November 4
Towne 337
Avik De
Anchor Synthesis via Template Composition
Click to view video of Avik De's seminar

Read the Abstract and Bio


We propose a new "synthetic" view of the template--anchor relation as an enhancement of the classical (analytical) view. Specifically, on top of merely being a relation between closed-loop systems, we introduce a reference plant, and a notion of mapping controllers from template to anchor. Additionally, we show that multiple templates can be anchored simultaneously on a robot body (parallel composition) and show two ways in which this modularity can be exploited. The example robot bodies at hand include a tailed hopping robot (Jerboa) and a quadruped with 8 actuated degrees of freedom (Minitaur).

Bio: Avik De completed his B.S./M.S. in Mechanical Engineering and a B.S. in Applied Math from Johns Hopkins University in 2010 and started a Phd program in ESE at Penn in Kodlab


November 11
F. Scott Stinner
Flexible, CdSe Nanocrystal Integrated Circuits
Click to view video F. Scott Stinner's seminar
Read the Abstract and Bio

Abstract: Solution and low-temperature processable semiconductors are being explored to realize low-cost, large-area, flexible electronics and to enable emerging mobile, wearable, and implantable devices. Applications including displays, sensors, integrated circuits (ICs), and radio frequency identification (RFID) systems have been demonstrated through the solution-based casting and printing of semiconducting organic molecules and polymers, carbon nanotube arrays, and sol-gel metal oxides. Recently, colloidal nanocrystals (NCs) have emerged as a new member of this family of solution-processable semiconductors with demonstrations of high mobility (>10 cm2/Vs) electronic transistors and integrated circuits.
I will be discussing the use of recently introduced methods of in-situ device repair allowing me to process CdSe NC materials, which are typically highly sensitive to their environment, outside of a nitrogen glovebox into the cleanroom where they are exposed to air and solvents commonly used in fabrication processes. I will discuss photolithographic methods used to pattern the device electrodes which allowed me to scale down device dimensions, drastically reducing parasitic capacitances, and scale up the fabrication to full 4 inch flexible Kapton substrates. I will be discussing NC field-effect transistors with channel lengths as small as 5 μm and electron mobilities up to 10 cm2/Vs. I will be discussing a newly developed process for creating VIA holes, allowing me to realize complex integrated circuits including inverters as well as NAND and NOR logic gates. Using building blocks, I am able to realize amplifiers with ~7 kHz bandwidth and ring oscillators with stage delays as low as 1.5 μs. 

Bio: F. Scott Stinner is a doctoral student in the Electrical and Systems Engineering Department at the University of Pennsylvania. Scott began his work at Penn in 2011 and earned his Master's in Electrical Engineering in 2013. Before coming to Penn, Scott earned a Bachelor's degree in Electrical and Computer Engineering from Lafayette College in Easton, PA. Scott's research interests include the development of large-scale deposition procedures for solution-based novel semiconducting materials, the design and fabrication of integrated circuits from novel semiconducting materials on flexible substrates, and the integration of novel semiconductor circuits into sensors. Scott's current research project focuses on integrating nanocrystal amplifiers with bio-sensors on plastic substrates. 


November 18
Wenxiang Chen
Optical devices based on nanocrystal plasmonic building blocks
Click to view Wenxiang Chen's seminar
Read the Abstract and Bio

Abstract:Nanoscale plasmonic devices offer an excellent ability to control light behavior in both near and far field. They can harness light’s energy for photovoltaics, focusing light for biosensing and tuning the phase of light for polarization optics. However, the fabrication for these devices usually depends on electron beam or focus ion beam lithography, which are slow and expensive. Here, we show that by nanoimprinting of gold colloidal nanocrystals, we can build large area nanoscale plasmonic structures in a fast and cheap way. Gold nanocrystal nano-antennas arrays and extreme bandwidth quarter-wave plates are demonstrated in experiments, and optically characterized to confirm their polarization ability. Quantum dot solar cells enhanced by plasmonic structures are also demonstrated by this method.

Wenxiang Chen is a doctoral student in Department of Electrical and Systems Engineering at the University of Pennsylvania from 2011. He received his bachelor degree in Applied Physics at the University of Science and Technology of China in 2011. His current work focuses on developing nanocrystal building blocks for plasmonic devices, such as nano-antennas, optical polarizers and plasmonic solar cells, and building plasmonic structures on flexible substrates.

  November 25
December 2
Mark Eisen
Quasi-newton Methods for Distributed Optimization
Click to view Mark Eisen's seminar
Read the Abstract and Bio

Abstract: In this work we adapt popular quasi-newton methods to solve convex optimization problems in a distributed setting, assuming both synchronous and asynchronous communication between nodes. Specifically, we examine the case in which each agent has a local variable and objective function and we wish to minimize the aggregate function while enforcing consensus across all nodes. We consider variations of the regularized BFGS and the limited memory BFGS that allow nodes to compute local functions and gradients using information available locally and with neighbors. These algorithms are defined and analyzed in the settings of synchronous communication as well as asynchronous communication, in which each nodes do not perform computations and send information on a common clock. We additionally provide formulations for both the primal and dual domains.

Mark Eisen received a Bachelor of Science in electrical engineering from the University of Pennsylvania, Philadelphia, USA in 2014. He is now working towards his PhD in the Department of Electrical and Systems Engineering at the University of Pennsylvania. His research interests include distributed optimization and machine learning. Mr. Eisen was awarded Outstanding Student Presentation at the 2014 Joint Mathematics Meeting, as well as the recipient of the 2014 Penn Engineering Exceptional Service Award.

December 9
Towne 337
Fernando Gama
Soft Clustering of Networked Data Using Cut Metrics
Click to view Fernando Gama's seminar
Read the Abstract and Bio

Abstract:The objective of clustering is to classify data into different classes following a given criteria to determine whether the data points are sufficiently alike or not. However, there are datasets that do not yield adequate results when data points are forced to belong to only one class. In order to overcome this difficulty, the notion of soft clustering arises. This allows a given data point to belong to more than one group. By using the concept of cut metrics, an algorithm for soft clustering is developed.

Bio: Fernando Gama is currently a doctoral student in the Department of Electrical and Systems Engineering at the University of Pennsylvania. His advisor is Alejandro Ribeiro. He is currently on a Fulbright Scholarship for International Students. He received his Electronic Engineering Degree from the University of Buenos Aires, Argentina in 2013. His research interests are in the Signal Processing field and its application to Network Theory.

December 16
Mohammad Hassan Lotfi
The Economics of Non-Neutrality on the Internet
Click to view Mohammed Hassan Lotfi's seminar
Read the Abstract and Bio

Abstract: Net neutrality on the Internet is perceived as the policy that mandates Internet Service Providers (ISPs) to treat all data equally, regardless of the source, destination, and type of the data.  It has almost been a decade that the advantages and disadvantages of the Internet non-neutrality have been put on debate. Proponents of Net-Neutrality claim that non-neutrality kills the innovation on the Content Providers (CP) side, decreases the competition among CPs, and undermines the so called ``free Internet". On the other hand, those who advocate relaxing the net-neutrality rules claim that the Internet neutrality, as it is perceived commonly, is a barrier to further developments on the Internet since it decreases the incentives of ISPs to invest on their infra-structure. This debate has mostly been conducted on a qualitative level without rigorous economic and technical analyzes. In this talk, I present an economic framework for the problem of non-neutrality adoption by ISPs in the presence of competition and CPs that can also be non-neutral, i.e. they can provide different quality of service over different ISPs. I discuss about the incentives of an individual ISP to migrate to a non-neutral regime, and the benefits of non-neutrality for ISPs in a setting in which some of the ISPs are neutral and some are non-neutral. 

 Mohammad Hassan Lotfi received the B.Sc. degree in electrical engineering from the Sharif University of Technology, Tehran, Iran, in 2011, and the M.S.c. degree in electrical and system engineering from the University of Pennsylvania, Philadelphia, PA, USA, in 2013, where he is currently pursuing the Ph.D. degree. His research interests are in economics of networks, game theory, pricing in communications and power networks, and net-neutrality.