ESE Colloquia & Events

Spring 2016

ESE colloquia are held on Tuesdays from 11-12:00pm in Towne 337, unless otherwise noted. For all Penn Engineering events, visit the Penn Calendar.

To be added to the ESE Events mailing list (which sends notifications regarding all departmental colloquia, seminars, and events) please email us at

Keren Bergman Tuesday, January 26
Alfred Hero
University of Michigan
Graph Continuum Limits in Data Science
Read the Abstract and Bio

Abstract: Many problems in data science fields including data mining, computer vision, and machine learning involve combinatorial optimization over a graphs, e.g., minimal spanning trees, traveling salesman tours, or k-point minimal graphs over a feature space.  Certain properties of minimal graphs like their length, minimal paths, or span have continuum limits  as the number of nodes approaches infinity. These include problems  arising in spectral clustering, statistical classification, multi-objective learning, and anomaly detection. In some cases these continuum limits lead to analytical approximations that can  break the combinatorial bottleneck.  In this talk, I will present an overview of some of the remarkable theory of graph continuum limits and illustrate with data science applications.

Bio: Alfred O. Hero III received the B.S. (summa cum laude) from Boston University (1980) and the Ph.D from Princeton University (1984), both in Electrical Engineering. Since 1984 he has been with the University of Michigan, Ann Arbor, where he is the R. Jamison and Betty Williams Professor of Engineering and co-director of the Michigan Institute for Data Science (MIDAS) . His primary appointment is in the Department of Electrical Engineering and Computer Science and he also has appointments, by courtesy, in the Department of Biomedical Engineering and the Department of Statistics. From 2008-2013 he held the Digiteo Chaire d'Excellence at the Ecole Superieure d'Electricite, Gif-sur-Yvette, France. He is a Fellow of the Institute of Electrical and Electronics Engineers (IEEE) and several of his research articles have received best paper awards. Alfred Hero was awarded the University of Michigan Distinguished Faculty Achievement Award (2011). He received the IEEE Signal Processing Society Meritorious Service Award (1998), the IEEE Third Millenium Medal (2000), and the IEEE Signal Processing Society Technical Achievement Award (2014). Alfred Hero was President of the IEEE Signal Processing Society (2006-2008) and was on the Board of Directors of the IEEE (2009-2011) where he served as Director of Division IX (Signals and Applications). He served on the IEEE TAB Nominations and Appointments Committee (2012-2014). Alfred Hero is currently a member of the Big Data Special Interest Group (SIG) of the IEEE Signal Processing Society. Since 2011 he has been a member of the Committee on Applied and Theoretical Statistics (CATS) of the US National Academies of Science.

Alfred Hero's recent research interests are in the data science of high dimensional spatio-temporal data, statistical signal processing, and machine learning. Of particular interest are applications to networks, including social networks, multi-modal sensing and tracking, database indexing and retrieval, imaging, biomedical signal processing, and biomolecular signal processing.

Keren Bergman

Thursday, February 4
Prineha Narang
11am, Moore 216
Light-matter Interactions at the Nanoscale: A Nonequilibrium Approach

Read the Abstract and Bio

Abstract:In this talk I will discuss a theory-driven approach, closely coupled with ultrafast spectroscopy and nanofabrication, aimed at the discovery of active optoelectronic materials and nanophotonic devices.

Surface plasmons, electromagnetic modes confined to the surface of a conductor-dielectric interface, have sparked recent interest because of their quantum nature and their broad range of applications. Despite more than a decade of intensive scientific exploration, new plasmonic phenomena continue to be discovered, including quantum interference of plasmons, observation of quantum coupling of plasmons to single particle excitations, and quantum confinement of plasmons in single-nm scale plasmonic particles. Simultaneously, plasmonic structures find widening applications in integrated nanophotonics, biosensing, photovoltaic devices, single photon transistors and single molecule spectroscopy. Decay of surface plasmons to hot carriers is a new direction that has attracted considerable fundamental and application interest, yet a detailed understanding of ultrafast plasmon decay processes and the underlying microscopic mechanisms remain incomplete.

In this talk I will provide a fundamental understanding of plasmon-driven hot carrier generation and relaxation dynamics in the ultrafast regime. I will report the first ab initio calculations of phonon-assisted optical excitations in metals as well as calculations of energy-dependent lifetimes and mean free paths of hot carriers, accounting for electron-electron and electron-phonon scattering, lending insight towards transport of plasmonically-generated carriers at the nanoscale. To conclude the first part of my talk, I will discuss recent experimental observations of the injection of these nonequilibrium carriers into molecules tethered to the metal surface and into wide bandgap nitride semiconductors.

In the second part of my talk I will present theory-directed design of Zn-IV nitride materials. The commercial prominence in the optoelectronics industry of tunable semiconductor alloy materials based on nitride semiconductor devices, specifically InGaN, motivates the search for earth-abundant alternatives for use in efficient, high-quality optoelectronic devices. II-IV-N2 compounds, which are closely related to the wurtzite-structured III-N semiconductors, have similar electronic and optical properties to InGaN namely direct band gaps, high quantum efficiencies and large optical absorption coefficients. The choice of different group II and group IV elements provides chemical diversity that can be exploited to tune the structural and electronic properties through the series of alloys. Here I will describe the first theoretical and experimental investigation of the ZnSnxGe1−xN2 series as a replacement for III-nitrides.

Finally I will give an outlook on the potential of excited state and non-equilibrium phenomena for nano-mesoscale devices.

Bio: Prineha received her Ph.D. in Applied Physics from the California Institute of Technology (Caltech), advised by Professors Harry A. Atwater and Nathan S. Lewis, as a National Science Foundation Graduate Fellow and Resnick Sustainability Institute Fellow. There she focused on light-matter interactions, ranging from quantum plasmonics to nitride optoelectronics. She was recently appointed as a Royal Society Newton Fellow to work in quantum photonics with Professor Sir John Pendry at Imperial College, London. Prineha’s research interests are in the area of excited state and ultrafast dynamics.


Monday, February 8
Aaswath Raman
Stanford University
10am, Singh Glandt Forum
Broadband Nanophotonics: Controlling Thermal Radiation and Light Absorption for Energy and Information Applications

Read the Abstract and Bio

Abstract: Meeting increasing global demand for energy while reducing carbon emissions remains one of the grand challenges of this century. Nanoscale photonic structures, by their small length scales, can manipulate light and heat in unprecedented ways, thereby enabling new possibilities for energy efficiency and generation to meet this challenge. In this talk, I will show how controlling the electromagnetic fields associated with thermal radiation and solar absorption using nanophotonic structures can fundamentally enable new technological capabilities for clean energy, by allowing us to better use both sunlight and another, unexploited renewable thermodynamic resource: the cold of space. Moreover, motivated by information applications, I will show how we can better characterize the fundamental behavior of important classes of nanophotonic devices over a broad range of wavelengths.

One particular application that motivates global interest in new approaches to energy efficiency is cooling, which is a significant end-use of energy globally and a major driver of peak electricity demand. I will present results of the first experimental demonstration of daytime radiative cooling, where a sky-facing nanophotonic surface passively achieved a temperature of 5-10°C below the ambient air temperature under direct sunlight. I will also discuss related work on using thermal nanophotonic approaches to passively maintain solar cells at lower temperatures, while maintaining their solar absorption, to improve their operating efficiency.

Motivated by the goal of better exploiting another source of renewable energy, sunlight, I will present a nanophotonic light trapping theory which reveals that, at the nanoscale, it is possible to exceed conventional limits on light trapping in solar cells for all absorption regimes, and explain the mechanisms for this enhancement. Finally I will introduce a plasmonic and metamaterial band theory that can rigorously model an important class of nanophotonic devices made of metallic or dispersive elements, over a broad range of wavelengths. This band theory offers insight into the performance of this class of devices in sensing and modulation applications.

Bio: Aaswath Raman is an Engineering Research Associate with the Ginzton Laboratory at Stanford University. He received his Ph.D. in Applied Physics from Stanford University in 2013, and his A.B. in Physics & Astronomy and M.S. in Computer Science from Harvard University in 2006. His research interests include nanophotonics, thermal science, renewable energy systems, solid-state devices and machine learning. In 2013, he was the recipient of the Stanford Postdoctoral Research Award, and in 2011, the SPIE Green Photonics Award, and the Sir James Lougheed Award of Distinction Fellowship from the Government of Alberta, Canada. In 2015, he received MIT Technology Review’s Innovator Under 35 (TR35) Award for being a Pioneer in Energy.

Thursday, February 11
Sam Emaminejad
UC, Berkeley
10am, Singh Glandt Forum
An Ecosystem of Integrated Physiological Monitoring Platforms for Personalized Medicine

Read the Abstract and Bio

Abstract: The number of interconnected sensors are expected to increase beyond trillions of units over the next decade, representing about 1000 devices per person in the world. The intersection of this trajectory with the pressing need for lowering healthcare expenditures necessitates a subset of these devices geared towards health monitoring of individuals to enable personalized medicine. This approach allows for the collection of large data sets which can guide clinical investigations and ultimately generate predictive algorithms to understand the clinical needs of individuals and society as a whole. In order to gain a comprehensive view of the health status of individuals, these devices should provide insight into the physiological biomarkers in humans’ samples.

To this end, I aim to develop an ecosystem of human biomonitoring platforms that can continuously analyze various humans’ physiological samples and their surrounding environments. In order to realize these platforms, an integrated-system approach must be adopted to accurately process and seamlessly relay the originated information from the sensor interface to cloud servers online. Furthermore, at the sensor-level, novel MEMs/NEMs-based technologies must be developed to perform actuation and sensing at length-scales comparable to the size of the target biomarkers. During this talk, I will discuss the design considerations that must be met to allow for extraction of meaningful information from physiological samples. In this context, I will present a flexible fully-integrated and wearable perspiration analyzer that accurately and simultaneously measures the main electrolytes (e.g. sodium and potassium) and metabolites (e.g. glucose and lactate) of sweat in real-time while calibrating the sensors' response against the change in skin temperature. Furthermore, I will present the unprecedented micro/nanoscale actuation and sensing functionalities that I demonstrated in the context of portable monitoring platforms.

I will conclude my talk with a discussion of future research directions which prelude my long term vision of developing an ecosystem of integrated portable, wearable, and implantable sensors to facilitate large scale population and epidemiological studies.

Bio: Sam Emaminejad received his BASc (2009) and MS/PhD (2011/2014) degrees in Electrical Engineering from the University of Waterloo and Stanford University, respectively. He pursued his PhD thesis at Stanford Genome Technology Center where he focused on applying micro- and nanotechnologies to develop low-cost and integrated biosensing and bioeletronics platforms. As a joint-postdoctoral scholar at UC Berkeley and Stanford School of Medicine, Sam is currently exploiting flexible electronics technology to develop non-invasive wearable sensors and systems for physiological monitoring and personalized medicine applications. Sam has previously worked as an ASIC and Analog Designer in semiconductor companies such as STMicroelectronics and Analog Devices. Sam was awarded Microsoft Merit and Natural Sciences and Engineering Research Council (NSERC) scholarships and was the recipient of the Best Paper Award at the IEEE Sensors conference in 2013. His current work has been widely reported by various media outlets including Nature, Science, Time, The Wall Street Journal, Newsweek, etc.

Thursday, February 18
Jun-Chau Chien
UC, Berkeley
11am, Moore 216
mm-Wave Lab-on-CMOS: Electromagnetic Sensing from Micro- to Nano-scales

Read the Abstract and Bio

Abstract: Lab-on-CMOS is an emerging platform for Point-of-Care diagnostics and precision medicine. By directly integrating active CMOS electronics into passive Lab-on-Chip devices, the new System-on-Chip not only offers ultimate device miniaturization but also enables highly integrated multiphysics biosensing and actuation. This research addresses the challenges in such a hybrid system while embracing the opportunities in system co-design to achieve improved sensing performance that leads to new scientific findings.

In this talk, I will present my research on a Lab-on-CMOS dielectric spectrometer for single-cell analysis using near-field sensing at millimeter-wave (mm-Wave) frequencies. The aim is to understand the wideband electromagnetic signatures at cellular and molecular levels and to open its way for real-time and label-free medical diagnostics and biological studies. I will focus on innovations in circuits, systems, microfluidics, and calibration techniques to enable a capacitance equivalent sensitivity limit of sub-aF, suitable for large-scale characterization of single-cell dielectric spectroscopy (6.5 ~ 30 GHz) in the setting of high-throughput flow cytometry. The capability of cell sorting based on frequency dispersion is demonstrated with the measurements of human breast cancer cells. In addition to electromagnetic sensing, I will introduce multiphysics actuation techniques to quantify the mechanical property of the cells. Specifically, the system measures the deformation of cells using hydrodynamic stretching. I will also discuss the challenges in sensing toward nano-scales and sub-THz frequencies and present an on-chip single-element electronic calibration (E-Cal) technique for nano-device measurements. In the end, I will conclude my talk with Lab-on-CMOS technology for new applications in sensing, imaging, and communication.

Bio: Jun-Chau Chien received the B.S. and M.S. degrees in Electrical Engineering from National Taiwan University in 2004 and 2006, respectively, and the Ph.D.
degree in Electrical Engineering and Computer Sciences from University of
California, Berkeley, in 2015. He is currently a post-doctoral research associate
at University of California, Berkeley. He has held industrial positions at
InvenSense, Xilinx, and HMicro working on mixed-signal integrated circuits for
inertial sensors and wireline/wireless transceivers. He is broadly interested in
innovative biotechnology for point-of-care diagnostics and medical imaging with
emphasis on silicon-based approaches.

Dr. Chien is the recipient of the 2006 Annual Best Thesis Award from Graduate
Institute of Electronics Engineering, National Taiwan University, the 2007
International Solid-State Circuit Conference (ISSCC) Silkroad Award, the co-
recipient of 2010 IEEE Jack Kilby Award for ISSCC Outstanding Student Paper,
the 2014 Analog Devices Outstanding Design Award, the 2014 Microwave
Theory and Techniques Society (MTT-S) Graduate Fellowship for Medical
Applications, the 2014 Solid-State Circuit Society (SSCS) Predoctoral
Achievement Award, and the 2014 UC Berkeley Outstanding Graduate Student
Instructor Award.

Tuesday, February 23
Minjie Chen
11am, Towne 337
Next Generation Power Electronics for High Impact Applications

Read the Abstract and Bio

Abstract: Power electronics is everywhere in our daily lives, with applications ranging from renewable energy and data centers to portable and biomedical electronics. These applications all require power electronics to offer high performance, small physical size, high reliability, and low cost. Traditionally, it has been preferable to implement power electronics with topologies that have low component count and simple controls. However, these designs require substantial energy storage and bulky passive components, and are reaching their fundamental limits with decreasing marginal performance gains. Moreover, they do not leverage the dramatic advances that have been made in semiconductor materials and integrated circuits. With the advent of wide-bandgap semiconductor materials, and the exciting opportunities offered by emerging high-impact applications, sophisticated and modularized power conversion architectures are becoming extremely attractive.

This talk will present three novel power electronics architectures developed for three important applications, including an electrolytic-free LED driver that promises substantially longer lifetime, a solar micro-inverter that delivers significantly higher efficiency, and a telecom power converter that realizes ultra-high power density. These three architectures extend the fundamental performance boundary of power electronics from three different perspectives, and are enlightening the path to much more sophisticated and modularized power electronics that will benefit a wide range of applications.

Bio: Minjie Chen received the B.S. degree from Tsinghua University in 2009, and the S.M., E.E., and Ph.D. degrees from MIT in 2012, 2014 and 2015, respectively. He is currently a postdoctoral research associate in the MIT Research Laboratory of Electronics. His primary research interests are in the design of high performance power electronics for emerging and high-impact applications, including renewable energy, lighting, grid-interface power electronics, and miniaturized power management systems. He is the recipient of the MIT E.E. Landsman Fellowship with a focus on power electronics, and a co- winner of an IEEE ECCE best student demonstration award.

RESCHEDULED to: Friday, February 26
Hui Fang
University of Illinois at Urbana-Champaign
11am, Berger Auditorium
Advanced Electronic Materials for Next-Generation Biomedical Implants and Bio‑tools

Read the Abstract and Bio

Abstract: Innovating electronic materials and related process technologies are critical in building next generation large scale, bio-electronic interface for biomedical implants and bio-tools.

In this talk, the influence of materials and process innovation will be discussed in the context of achieving two essential properties at the bio-electronic interface, bio‑conformality and bio‑stability.  First, to reconcile the mechanic properties mismatch between soft, curvilinear organ surface and conventional rigid, planar electronics, Si nanomembrane enables bio‑conformal electronics from top down approach and advanced micro/nano fabrication on flexible substrates.  In the second part of the talk, I will discuss how to achieve long term bio-stability at the bio-electronic interface through an ultrathin hermetic thermal silicon dioxide layer from a special device fabrication process.  A capacitively coupled, bio-conformal sensing electronics with over 1,000 channels demonstrate the robustness of this encapsulation strategy.  Together, these results form a realistic pathway towards bio‑compatible, bio‑conformal and bio‑stable electronic implants, with potential for broad utility, such as brain/heart activity mapping, brain-machine interface, and pharmaceutical screening.  At the end of my talk, I will show how we can leverage recent advancements in nano‑electronics into building next generation bio‑electronics and solve big problems in biology, especially in brain activity mapping.

Bio: Dr. Hui Fang received his B.S. degree (2009) from Tsinghua University, and Ph.D. degree (2014) from the University of California, Berkeley, both in Materials Science and Engineering.  At Berkeley he worked under the supervision of Prof. Ali Javey.  Currently Dr. Fang is a postdoctoral research associate in Professor John A. Rogers’ group at the University of Illinois, Urbana‑Champaign. Dr. Fang’s research interests include developing novel materials, devices and related process technologies for bio-integrated electronics and nano‑electronics, as well as exploring new materials/device physics at the nanoscale.  Dr. Fang is a recipient of the 2013 Chinese Government Award for Outstanding Self-financed Students Abroad. 

His publication record can be found online at

Tuesday, March 1
Kejie Fang
11am, Towne 337
Integrated Hybrid Photonics: Emergent Control and Application of Light and Sound at Nanoscale

Read the Abstract and Bio

Abstract: The bottleneck of bandwidth limitation and power dissipation in today's electronic microchips is conflicting with the exceeding demand for information communication and processing. Light, due to its intrinsic high frequency and environment-insensitivity (owing to its charge neutrality), is expected to bring solutions to this fundamental challenge. By the same token, certain functionalities in optical information processing will require a hybrid architecture interfacing different materials and light-matter interactions. With technical advances in nanofabrication, it is now possible to manipulate light and enhance light-matter interactions in on-chip, nanoscale photonic structures. In this talk, I will present my research in two integrated hybrid photonic architectures. First is optoelectronic integration, where we achieved novel active control of light through an electric drive which dynamically modulates the refractive index of silicon photonic structures, leading to an effective magnetic field for photons and topological light propagation. These novel interactions are unreachable in static or passive dielectrics and provide a solution for on-chip optical isolation that is essential for stable and energy efficient optical communication. In the second part of my talk, I will present work on another hybrid architecture that interfaces light and sound: optomechanical crystals. This architecture allows for simultaneously engineering of optical and mechanical properties as well as photon-phonon interactions. Combining electron beam lithography and scanning probe microscope tuning, we fabricated cavity-optomechanical circuits on silicon microchips to realize radiation-pressure controlled microwave phonon routing. We applied these devices for microwave-over-optical signal processing with low energy and high efficiency. The nanoscale mechanical vibration is also used to achieve optical non-reciprocity in the optomechanical circuit. These achievements hold promise for hybrid photonic technology for light-based communication and processing in an integrated, chip-scale platform.

Bio: Kejie Fang is a Postdoctoral Scholar in Applied Physics at California Institute of Technology, working with Prof. Oskar Painter. He received his B.S. in physics from Peking University, and his M.S. in electrical engineering, Ph.D. in physics, both from Stanford University under the supervision of Prof. Shanhui Fan. Kejie's research interests include optomechanics, nanophotonics, and spin photonics, with a theme to develop novel chip scale devices and systems for light-based applications including optical information communication and processing. During his Ph.D., he proposed and demonstrated for the first time an effective magnetic field for photons which provides a solution for on-chip optical isolation. At Caltech, he developed integrated cavity-optomechanical circuits for on-chip information processing using nanoscale optical and acoustic excitations. Kejie has published 15 peer-reviewed papers in leading journals including Nature Photonics, Physical Review Letters, and Nature Communications. Kejie was a William R. and Sara Hart Kimball Fellow at Stanford University and also a recipient of OSA Outstanding Reviewer Award in 2014.

Wednesday, March 2
Emmanuel Abbe
Princeton University
11am, Towne 337
The Quest of Fundamental Limits in Community Detection

Read the Abstract and Bio

Abstract: Community detection is at the heart of network and data sciences, with a broad range of applications in engineering, social and natural sciences. While various methodologies have expanded rapidly, understanding which community structure can or cannot be extracted has long remained a challenge. Recently, this picture has changed due to new developments on statistical network models. This talk overviews our contributions to this field, in particular the establishment of fundamental limits for block models, and the new insights that emerge for clustering algorithms. It also points out new directions to pursue for a foundational approach to graph mining.

Bio: Emmanuel Abbe is an assistant professor at Princeton University, jointly in the Program for Applied and Computational Mathematics and the Department of Electrical Engineering. He previously received his Ph.D. from the EECS Department at MIT and his M.S. from the Mathematics Department at EPFL. His research interests are in statistical networks, information theory, learning theory, and applied probability. He received the 2011 Foundation Latsis International Prize, the 2014 Bell Labs Prize, the NSF CAREER Award and the 2016 Google Faculty Research Award.

Thursday, March 3
Ethem Erkan Aktakka
University of Michigan
11am, Moore 216
Toward Self-Powered, Self-Calibrated and Multifunctional Smart Microsystems

Read the Abstract and Bio

Abstract: The world is evolving to become more instrumented, interconnected and intelligent with an influx of wireless microsystems into every aspect of our lives, from wearable electronics and bio-implants to smart infrastructures and industrial/environmental monitoring systems. The greatest impediment to global deployment of the next generation wireless microsystems is the need for autonomous and reliable operation over a long-term unattended use. In this regard, tethering to a power outlet or a finite-lifetime battery limits the application field, and also results up to a tenfold markup in the cost of installation and maintenance. In addition, the utilized sensors can experience drifts in their output signals over the course of usage due to various environmental and device specific factors, often enough to require frequent replacements and restrict their potential in high accuracy applications. In the first part of this talk, I will outline my research on micro energy harvesting technologies to enable truly wireless and energy-independent systems, which harness vibrational, thermal, or acoustic power in the environment. Challenges of achieving high efficiency, wide bandwidth, multi-axis operation, and CMOS integration will be addressed. Some of the devices to be demonstrated include a 200-μW-output micro vibration harvester packaged with its self-supplied power management IC, and an on-chip thermal energy harvester integrated on a FinFET CMOS substrate. To realize these devices, new technology platforms are developed, including low-temperature microfabrication of high quality piezoelectric films on silicon, and co-evaporation of bismuth/antimony telluride thermoelectric thin films. The second part of the talk will focus on enhancing the performance of next generation sensors by integrating self-calibration and vibration isolation functionalities via micro mechatronic systems. A six degrees-of-freedom micro motion stage will be demonstrated to provide precise on-chip physical stimulus to a multi-axis inertial sensor for in situ measurement and recalibration of its signal drifts. The talk will conclude with discussion of future opportunities for smart multi-functional microsystems and self-adaptive energy conversion technologies.

Bio: Erkan Aktakka is a senior research fellow in the Department of Electrical and Computer Engineering at the University of Michigan. He received his B.S. degree from the Middle East Technical University in Turkey in 2006, and the M.S. and Ph.D. degrees while working with Prof. Khalil Najafi at the University of Michigan, Ann Arbor in 2008 and 2012, respectively, all in electrical engineering. Dr. Aktakka is a recipient of the 1st place in Turkey’s national university entrance exam, TUBITAK Graduate Research Fellowship Award, DTE Clean Energy Prize, and University of Michigan Distinguished Achievement Award. His research interests are in micro/nano electro-mechanical systems, smart materials and structures, energy harvesting, acoustic/ultrasonic transducers, and multi-functional microsystems.

Friday, March 4
Azalia Mirhoseini 
University of California, San Diego
10am, Towne 337
Bringing the Machine into the Loop of Machine Learning

Read the Abstract and Bio

Abstract: Contemporary analytical algorithms are often focused on functionality and accuracy with system performance as an afterthought. As their use/scale grows and the computing platforms become diverse, spanning from servers and desktops to smartphones and Internet of Things (IoT) devices, functionality is not just about algorithmic efficiency and accuracy, but also practicality on real-world computing machines. One-size fits all solutions will not meet the physical needs of emerging analytical application scenarios.

In this talk, I will present my research on novel computing frameworks that bring hardware into the loop of designing scalable inference algorithms and learning systems. I will describe how a multi-faceted design that holistically considers the computing domain parameters, namely data, algorithm, and machine, introduces game changing performance gains across the board, including runtime, energy, memory, and network bandwidth. I will then describe my tools which enable automatic end-to-end adoption of the proposed frameworks in a wide range of data inference application scenarios. On the theoretical side, I show how my new solutions reach the target machine's computation/communication bounds. On the practical side, I present customized approaches for a range of algorithms and applications (e.g., penalized regression, classification, and deep neural networks), datasets (e.g., visual and sensing), and machines (e.g., GPU, FPGA, CPU clusters, and heterogeneous architectures). I also demonstrate my approach towards enabling single-pass streaming learning problems. Finally, I discuss how lessons learned in the context of my holistic frameworks can bring new directions in the design of broader analytical scenarios such as privacy preserving and just-in-time computing

Bio: Azalia Mirhoseini is a postdoctoral researcher in the department of Electrical and Computer Engineering (ECE) at the University of California, San Diego and Rice University. She received her Ph.D. from Rice University where she worked on algorithms and architectures for performance efficient data analytics. Azalia has received multiple awards, including best 2015 Ph.D. thesis award at Rice ECE department, gold medal in the national math Olympiad in Iran, and fellowships from IBM, Schlumberger, and Microsoft Research.

Thursday, March 10
Saman Saeedi
11am, Moore 216
Holistic Design in High-Speed Optical Interconnects

Read the Abstract and Bio

Abstract: Integrated circuit scaling has enabled a huge growth in processing capability, which necessitates a corresponding increase in inter-chip communication bandwidth. As bandwidth requirements for chip-to-chip interconnection scale, deficiencies of electrical channels become more severe. Optical links present a viable alternative due to their low frequency-dependent loss and higher bandwidth density in form of WDM. Increasing silicon integration leads to better performance in optical links but necessitates a corresponding intimate design strategy of electronics and photonics. In this light, designing optical links with a deep understanding of photonics and state-of-the-art electronics brings their performance to an unprecedented level.

In this talk a 3D-integrated CMOS/Silicon-photonic receiver will be presented. This receiver is designed to effectively take advantage of low-cap silicon photonic photodiodes and advanced 3D-integration technologies. The electronic chip features an integrating receiver based on a low-bandwidth TIA that employs double-sampling and equalization through dynamic offset modulation. This architecture is also implemented in a 4-channel WDM-based parallel optical receiver using a forwarded clock at quarter-rate. Quadrature ILO-based clocking is employed for synchronization and a novel frequency-tracking method that exploits the dynamics of IL in a quadrature ring oscillator to increase the effective locking range.

Next, I will describe some of my research in other primary elements of an optical link: clocking and transmitters. A first-order frequency synthesizer will be presented that is suitable for high-speed optical links as well as on-chip clock generation. This frequency synthesizer is capable of receiving a low jitter optical reference clock generated by a high-repetition-rate mode-locked laser. On the optical transmitter side, two new techniques will be presented. First technique is thermal stabilization of micro-ring resonator modulators through direct measurement of ring temperature using a monolithic PTAT temperature sensor. Second technique is a differential ring modulator that breaks the optical bandwidth/quality factor trade-off known to limit the speed of high-Q ring modulators. This structure maintains a constant energy in the ring to avoid pattern-dependent power droop.

Lastly, I will cover some future directions for my research beyond data communication networks. Low-power mixed-signal circuit design has the potential to enable electronic neuromorphic neural networks that can scale to biological levels. Such artificial neural network would be used to build robots whose intelligence matches that of mice. On another front, miniature implantable microsystems can be deployed in the human body for constant monitoring of the vital signs and blood components such as oxygen and glucose. Implantable devices are also finding their way in diagnosis, treatment, and monitoring neurological disorders such as epilepsy, depression, and Parkinson’s disease. Leveraging low power mixed-signal circuit design techniques such miniaturized systems can be implemented in advanced CMOS technologies.

Bio: Saman Saeedi received his double-major B.S. degree in Electrical Engineering and Physics from Sharif University of Technology, Tehran, Iran, in 2010. He received his M.S. and Ph.D. degrees in Electrical Engineering from California Institute of Technology, Pasadena, in 2011 and 2015 respectively. He is currently a member of VLSI research group at Oracle Labs. The focus of his current research is low-power, high-performance mixed-signal integrated circuits with applications in sensing and communication. During the summer of 2012 he was a PhD intern at Apple Inc. where he worked on display driver chipsets. His work during fall and winter of 2014 at Rockley Photonics Inc. enabled a core technology for CMOS/silicon-photonic optical packet switching in data centers.

Dr. Saeedi is a gold medal winner of the National Physics Olympiad and recipient of four years undergraduate fellowship from National Elite Foundation of Iran. He received the Atwood fellowship in Fall 2010 and is the recipient of 2014 Intel/Texas Instruments/Catalyst Foundation CICC Student Scholarship Award and a finalist of 2015 Broadcom Foundation University Research Competition. Dr. Saeedi holds 7 U.S. patents and is serving on the Technical Program Committee of IEEE Optical Interconnect Conference.

Friday, March 11
Guanghua Shu
University of Illinois at Urbana-Champaign
11am, Moore 216
Towards Energy Proportional Communication Links
Read the Abstract and Bio

Abstract: With the explosive growth of data traffic in the Big Data era, intra- and inter-chip data communication links in both high performance computing systems and mobile platforms are starting to consume significant fraction of the system power. There have been many efforts to aggressively improve the energy efficiency of wireline links. Such efforts have mainly focused on improving energy efficiency of the link building blocks operating at peak performance. Unfortunately, these improvements do not necessarily translate to energy savings at the system level, because links are seldom fully utilized at peak performance. In other words, energy efficiency degrades drastically in many practical applications where links are only sporadically utilized.

Major part of this talk explores system-level opportunities to improve link energy efficiency. Based on the observation that real world data traffic exhibits bursty behavior with intermittent active and idle periods, I will first discuss techniques that seek to minimize power wastage during the idle periods and achieve energy proportional operation. I will then present techniques to achieve better than energy proportional operation by performing dynamic voltage and frequency scaling (DVFS). Circuit implementation details including integrated power management techniques used in the prototype transceiver to achieve less than 14ns wake up time and 500x (8Gb/s to 16Mb/s) energy proportional range will be described, along with the experimental results.

I will conclude with some thoughts on extending the concept of energy proportionality to not only other link topologies, but also many interface technologies in general.

Bio: Guanghua Shu is currently a Ph.D. candidate in the Department of Electrical and Computer Engineering at University of Illinois at Urbana-Champaign, IL, USA. He received M.S. degree in Microelectronics from Fudan University, Shanghai, China, in 2011. In the summer of 2014, he was a Research Intern in Xilinx, San Jose, CA, developing power and area-efficient parallel link architectures. In the fall of 2014 and summer of 2015, he worked on 56Gb/s wireline receivers (both electrical and optical) in Mixed-Signal Communication IC Design group in IBM Thomas J. Watson Research Center, Yorktown Heights, NY. His research interests include energy-efficient wireline communication systems, clocking circuits, power converters, and ultra-low-power circuits and systems for biomedical applications.

He is a recipient of University of Illinois Dissertation Completion Fellowship (2015-2016), and the IEEE SSCS Predoctoral Achievement Award (2014-2015).

Tuesday, March 15
Carlee Joe-Wong
Princeton University
11am, Towne 337
Smart Data Pricing

Read the Abstract and Bio

Abstract: Data traffic has increased sharply over the past decade and is expected to grow further as the Internet becomes ever more popular. Yet data network capacity is not expanding fast enough to handle this exponential growth, leading service providers to change their mobile data plans in an effort to reduce congestion. Inspired by these ongoing changes and building on work from the 1990s, smart data pricing (SDP) aims to rethink data pricing fortomorrow'snetworks. In this talk, I will focus on first the temporal and then the content dimensions of SDP. Time-dependent pricing (TDP) proposes to lower short-lived peaks in network congestion by incentivizing users to shift their data usage to less congested times. While TDP has been used in industries such as smart grids, TDP for mobile data presents unique challenges, e.g., it is difficult to predict how users will react to the prices on different days. Thus, we developed algorithms that continually infer users' changing responses to the offered prices, without collecting private data usage information. We implemented these algorithms in a prototype system, which we used to conduct the first field trial of TDP for mobile data. We showed that our TDP algorithms led to significantly less temporal fluctuation in demand, benefiting the service provider and lowering users' data prices overall.

Sponsored data, an emerging form of data pricing offered by AT&T, allows content providers to subsidize their users' data traffic; the resulting revenue can be used to expand existing data networks. We consider the impact of sponsored data on different content providers and users, showing that cost-aware users and cost-unaware content providers reap disproportionate benefits. Simulations across representative users and content providers verify that sponsored data may help to bridge the digital divide between different types of users, yet can exacerbate competition between content providers.

Bio: Carlee Joe-Wong is a Ph.D. candidate and Jacobus fellow at Princeton University's Program in Applied and Computational Mathematics. She is interested in mathematical aspects of computer and information networks, including work on smart data pricing and fair resource allocation. Carlee received her A.B. in mathematics in 2011 and her M.A. in applied mathematics in 2013, both from Princeton University. In 2013–2014, she was the Director of Advanced Research at DataMi, a startup she co-founded from her data pricing research. Carlee received the INFORMS ISS Design Science Award in 2014 and the Best Paper Award at IEEE INFOCOM 2012. She was a National Defense Science and Engineering Graduate Fellow (NDSEG) from 2011 to 2013.

Thursday, March 17
Dimitris Papailiopoulos
UC, Berkeley
11am, Moore 216
Less Talking, More Learning: Avoiding Coordination in Parallel Machine Learning Algorithms

Read the Abstract and Bio

Abstract: The recent success of machine learning (ML) in both science and industry has generated an increasing demand to support ML algorithms at scale. In this talk, I will discuss strategies to gracefully scale machine learning on modern parallel computational platforms. A common approach to such scaling is coordination-free parallel algorithms, where individual processors run independently without communication, thus maximizing the time they compute. However, analyzing the performance of these algorithms can be challenging, as they often introduce race conditions and synchronization problems.

In this talk, I will introduce a general methodology for analyzing asynchronous parallel algorithms. The key idea is to model the effects of core asynchrony as noise in the algorithmic input.  This allows us to understand the performance of several popular asynchronous machine learning approaches, and to determine when asynchrony effects might overwhelm them.  To overcome these effects, I will propose a new framework for parallelizing ML algorithms, where all memory conflicts and race conditions can be completely avoided. I will discuss the implementation of these ideas in practice, and demonstrate that they outperform the state-of-the-art across a large number of ML tasks on gigabyte-scale data sets.

Bio: Dimitris Papailiopoulos is a postdoctoral researcher in the Department of Electrical Engineering and Computer Sciences at UC Berkeley and a member of the AMPLab. His research interests span machine learning, coding theory, and parallel and distributed algorithms, with a current focus on coordination-free parallel machine learning, large-scale data and graph analytics, and the use of codes to speed up distributed computation. Dimitris completed his Ph.D. in electrical and computer engineering at UT Austin in 2014. At Austin he worked under the supervision of Alex Dimakis. In 2015, he received the IEEE Signal Processing Society, Young Author Best Paper Award.

Tuesday, March 29
Hamed Hassani
ETH Zurich
11am, Towne 337
From Communication to Sensing and Learning: An Information Theoretic Perspective

Read the Abstract and Bio

Abstract: We are witnessing a new era of science — ushered in by our ability to collect massive amounts of data and by unprecedented ways to learn about the physical world. Beyond the challenges of storage and communication, there are new frontiers in the acquisition, analysis and exploration of data. In this talk, I will view these frontiers through the lens of information theory. I will argue that information theory lies at the center of data science, offering insights beyond its classical applications. As a concrete example, I will consider the problem of optimal data acquisition, a challenge that arises in active learning, optimal sensing and experimental design. Based on information theoretic foundations, and equipped with tools from submodular optimization theory, I will present a rigorous analysis of the widely-used sequential information maximization policy (also known as the information-gain heuristic). Our analysis establishes conditions under which this policy provably works near-optimally and identifies situations where the policy fails. In the latter case, our framework suggests novel, efficient surrogate objectives and algorithms that outperform classical techniques.

Bio: Hamed Hassani is a post-doctoral scholar at the Institute for Machine Learning at ETH Zurich. He received a Ph.D. degree in Computer and Communication Sciences from EPFL, Lausanne. Prior to that, he received a B.Sc. degree in Electrical Engineering and a B.Sc. degree in Mathematics from Sharif University of Technology, Tehran. Hamed's fields of interest include machine learning, coding and information theory as well as theory and applications of graphical models. He is the recipient of the 2014 IEEE Information Theory Society Thomas M. Cover Dissertation Award. His co-authored paper at NIPS 2015 was selected for an oral (plenary) presentation, and his co-authored paper at ISIT 2015 received the IEEE Jack Keil Wolf ISIT Student Paper Award.

Wednesday, March 30 - RESCHEDULED
Mert Gürbüzbalaban
11am, Towne 337
Analyzing Complex Systems and Networks: Incremental Optimization and Robustness

Read the Abstract and Bio

Abstract: Many of the emergent technologies and systems including infrastructure systems (communication, transportation and energy systems) and decision networks (sensor and robotic networks) require rapid processing of large data and comprise dynamic interactions that necessitate robustness to small errors, disturbances or outliers.

Motivated by large-scale data processing in such systems, we first consider additive cost convex optimization problems (where each component function of the sum represents the loss associated to a data block) and propose and analyze novel incremental gradient algorithms which process component functions sequentially and one at a time, thus avoiding costly computation of a full gradient step. We focus on two randomized incremental methods, Stochastic Gradient Descent (SGD) and Random Reshuffling (RR), which have been the most widely used optimization methods in machine learning practice since the fifties. The only difference between these two methods is that RR samples the component functions without-replacement whereas SGD samples with-replacement. Much empirical evidence suggested that RR is faster than SGD, although no successful attempt has been made to explain and quantify this discrepancy for a long time. We provide the first theoretical convergence rate result ofO(1/k2s) for any s in (1/2,1) (andO(1/k2) for a bias-removed novel variant) with probability one for RR showing its improvement overΩ(1/k) rate of SGD and highlighting the mechanism for this improvement. Our result relies on a detailed analysis of deterministic incremental methods and a careful study of random gradient errors. We then consider deterministic incremental gradient methods with memory and show that they can achieve a much-improved linear rate using a delayed dynamical system analysis.

In the second part, we focus on large-scale continuous-time and discrete-time linear dynamical systems that model various interactions over complex networks and systems. There are a number of different metrics that can be used to quantify the robustness of such dynamical systems with respect to input disturbance, noise or error. Some key metrics are the H-infinity norm and the stability radius of the transfer matrix associated to the system. Algorithms to compute these metrics exist, but they are impractical for large-scale complex networks or systems where the dimension is large and they do not exploit the sparsity patterns in the network structure. We develop and analyze the convergence of a novel scalable algorithm to approximate both of the metrics for large-scale sparse networks. We then illustrate the performance of our method on numerical examples and discuss applications to design optimal control policies for dynamics over complex networks and systems.

Bio: Mert Gürbüzbalaban is a postdoctoral associate at the Laboratory for Information and Decision Systems (LIDS) at MIT. He is broadly interested in optimization and computational science driven by applications in large-scale information and decision systems and networks. Previously, he received his B.Sc. degrees in Electrical Engineering and Mathematics as a valedictorian from Boğaziçi University, Istanbul, Turkey, the “Diplôme d’ingénieur” degree from École Polytechnique, France, and the M.S. and Ph.D. degrees in Applied Mathematics from the Courant Institute of Mathematical Sciences, New York University.

Dr. Gürbüzbalaban received the Kurt Friedrichs Prize (given by the Courant Institute of New York University for an outstanding thesis) in 2013, Bronze Medal in the École Polytechnique Scientific Project Competition in 2006, the Nadir Orhan Bengisu Award (given by the electrical-electronics engineering department of Boğaziçi University to the best graduating undergraduate student) in 2005 and the Bülent Kerim Altay Award from the Electrical-Electronics Engineering Department of Middle East Technical University in 2001.

Wednesday, April 6
Varun Jog
Warren Center for Network and Data Sciences, and the Department of Statistics, University of Pennsylvania
11am, Towne 337
Modeling Stochastic Networks: Sharp Thresholds and Persistent Phenomena

Read the Abstract and Bio

Abstract: Recent years have seen a proliferation of large datasets possessing some type of network structure. Various generative models have been derived to reflect mathematical properties of real-world networks, including degree distributions and levels of connectivity and clustering. In the first part of the talk, we will discuss recent results in community detection and estimation. This is a very active area of research in electrical engineering and statistics, with diverse scientific applications in physics, biology, and social and behavioral sciences. However, the case where the observed adjacency matrix is "weighted" or "colored" has received relatively little attention. We will present sharp thresholds for when community detection is possible in weighted stochastic block models (SBMs), stated in terms of an information-theoretic expression known as the Renyi divergence. We will then discuss algorithms for recovering communities in an efficient manner in the regime when recovery is possible. In the second part of the talk, we will shift our focus to the problem of studying node centrality in growing random graphs. Many notions of graph centrality have been formulated in network science to assign numerical measures of importance to nodes in a graph. As the graph evolves dynamically over time, however, the central node(s) may also change. We will show that for particular classes of growing random graphs, namely preferential and uniform attachment trees, a particular notion of centrality persists -- meaning the most central node(s) settle down after finitely many steps. Furthermore, we will show how to guarantee this property for the first node if it seeds the graph with sufficiently many neighbors. We will conclude by discussing interesting challenges and open questions in network science and engineering.

Bio: Varun Jog is a postdoctoral fellow at the Warren Center for Network and Data Sciences and the Statistics department at the University of Pennsylvania. He received his B.Tech. degree in Electrical Engineering from IIT Bombay (2010), and his PhD in Electrical Engineering & Computer Sciences (EECS) from UC Berkeley (2015), where he was advised by Prof. Venkat Anantharam. His research interests include information theory, network science, energy harvesting, convex geometry, and optimal transport. He is a recipient of the Eliahu Jury award from the EECS Department at UC Berkeley (2015) and the Jack Wolf student paper award at the IEEE International Symposium on Information Theory (2015).

Wednesday, June 15th - ESE Seminar
Yannis Paschalidis
Boston University
11:30am, Towne 337
Data-Driven Model Estimation from Observed Equilibria: What transportation networks and bacterial cells have in common

Read the Abstract and Bio

Abstract: Equilibrium modeling is common in a variety of fields such as game theory, transportation science, and cell metabolism. The inputs to these models, however, are often difficult to estimate, while their outputs, i.e., the equilibria they are meant to describe, are often directly observable. By combining ideas from inverse optimization with the theory of variational inequalities, I will present an efficient, data-driven technique for estimating the parameters of these models from observed equilibria. Our framework allows for both parametric and non-parametric estimation and provides probabilistic guarantees on the quality of the estimated quantities.

I will present applications in two seemingly distinct areas. In transportation networks we use these techniques to estimate the congestion function, which determines users' route selection. Using actual traffic data from the Boston area, we can assess the inefficiency of drivers' selfish behavior vs. a socially optimal solution, also called the price of anarchy.

In biochemical networks, Flux Balance Analysis (FBA) is a widely used predictive model which computes a cell's steady-state chemical reaction fluxes as a solution to an optimization problem. FBA, however, assumes a certain global cellular objective function which is not necessarily known. We will use our new method to estimate such an objective. This enables us to elucidate the cellular metabolic control mechanisms and infer important information regarding an organism's evolution.

Bio: Yannis Paschalidis is a Professor and Distinguished Faculty Fellow of Electrical and Computer Engineering, Systems Engineering, and Biomedical Engineering at Boston University. He is the Director of the Center for Information and Systems Engineering (CISE). He obtained a Diploma (1991) from the National Technical University of Athens, Greece, and an M.S. (1993) and a Ph.D. (1996) from the Massachusetts Institute of Technology (MIT), all in Electrical Engineering and Computer Science. He has been at Boston University since 1996. His current research interests lie in the fields of systems and control, networks, optimization, operations research, computational biology, and medical informatics.

Prof. Paschalidis' work on communication and sensor networks has been recognized with a CAREER award (2000) from the National Science Foundation, the second prize in the 1997 George E. Nicholson paper competition by INFORMS, the best student paper award at the 9th Intl. Symposium of Modeling and Optimization in Mobile, Ad Hoc, and Wireless Networks (WiOpt 2011) won by one of his Ph.D. students for a joint paper, and an IBM/IEEE Smarter Planet Challenge Award. His work on protein docking (with his collaborators) has been recognized for best performance in modeling selected protein-protein complexes against 64 other predictor groups (2009 Protein Interaction Evaluation Meeting). His recent work on health informatics won an IEEE Computer Society Crowd Sourcing Prize.

He was an invited participant at the 2002 Frontiers of Engineering Symposium organized by the National Academy of Engineering, and at the 2014 National Academies Keck Futures Initiative (NAFKI) Conference. Prof. Paschalidis is a Fellow of the IEEE and the Editor-in-Chief of the IEEE Transactions on Control of Network Systems.