Scientific Computing and Matrix Computations Seminar: Why Deep Learning Works: Traditional and Heavy-Tailed Implicit Self-Regularization in Deep Neural Networks
Seminar: Scientific Computing: CS | May 1 | 12-1 p.m. | 380 Soda Hall
Michael W. Mahoney, ICSI and Department of Statistics, University of California at Berkeley
Random Matrix Theory (RMT) is applied to analyze the weight matrices of Deep Neural Networks (DNNs), including both production quality, pre-trained models and smaller models trained from scratch. Empirical and theoretical results clearly indicate that the DNN training process itself implicitly implements a form of self-regularization, implicitly sculpting a more regularized energy or penalty... More >
Seminar: Dissertation Talk: EE | May 1 | 3-4 p.m. | 380 Soda Hall
Stephen Tu, UC Berkeley
Reinforcement learning (RL) has demonstrated impressive performance in various domains such as video games, Go, robotic locomotion, and manipulation tasks. As we turn towards RL to power autonomous systems in the physical world, a natural question to ask is, how do we ensure that the behavior observed in the laboratory reflects the behavior that occurs when systems are deployed in the real world?... More >
Rapidly mixing random walks on matroids and related objectsidly mixing random walks on matroids and related objects
Seminar: CS | May 1 | 3-4 p.m. | 1011 Evans Hall
Nima Anari, Stanford University
A central question in randomized algorithm design is what kind of distributions can we sample from efficiently? On the continuous side, uniform distributions over convex sets and more generally log-concave distributions constitute the main tractable class. We will build a parallel theory on the discrete side, that yields tractability for a large class of discrete distributions. We will use this... More >
Center for Computational Biology Seminar: Sohini Ramachandran, Associate Professor, Brown University
Seminar: Biosystems and Computational Biology: CS | May 1 | 4:30-5:30 p.m. | 125 Li Ka Shing Center
Leveraging linkage disequilibrium to identify adaptive and disease-causing mutations
Correlation among genotypes in human population-genetic datasets complicates the localization of both adaptive mutations and disease-causing mutations. I will describe our latest efforts to develop new methods for localizing adaptive and disease-causing mutations, motivated by (1) incorporating... More >
Seminar: Dissertation Talk: EE | May 2 | 2-3 p.m. | 606 Soda Hall
Frank Li, UC Berkeley
Seminar: Dissertation Talk: CS | May 3 | 11 a.m.-12 p.m. | 465H Soda Hall
In this thesis, we argue that serverless functions represent a viable platform for analytics workloads, eliminating cluster management overhead, fulfilling the promise of elasticity. We identify and provide solutions to address two major issues in realizing this vision.
Solid State Technology and Devices Seminar: 24/7 Electricity Produced by Intermittent Power Requires Its Energy Storage
Seminar: Solid State Technology and Devices: EE: CS | May 3 | 1-2 p.m. | Cory Hall, The Hogan Room, 521
Jerry Woodall, Electrical and Computer Engineering, UC Davis
This is a simple story with a no-brainer punchline included in the title. Except for geothermal and nuclear energy, the sun is, and has been, the source of nearly all energy used on our planet. The problem is that the earth receives plenty of intermittent solar power, but not as solar energy.
Seminar: Dissertation Talk: CS | May 3 | 1:30-2:30 p.m. | 306 Soda Hall
Jingcheng Liu, UC Berkeley
In classical statistical physics, a phase transition is understood by studying the geometry (the zero-set) of an associated polynomial (the partition function). In this talk I will show that one can exploit this notion of phase transitions algorithmically, and conversely exploit the analysis of algorithms to understand phase transitions. As applications, I will give efficient deterministic... More >
Seminar | May 3 | 2-3 p.m. | 4 LeConte Hall
Watch a brief animated explainer of bionanotechnology at http://www.shawndouglas.com
Seminar: Dissertation Talk: EE | May 3 | 2:30-3:30 p.m. | 531 Cory Hall
Ankush Pankaj Desai, University Of California, Berkeley
Asynchronous event-driven systems can be found in myriad domains including cloud computing systems, device drivers, and robotics.
These systems are notoriously hard to get right as the programmer needs to reason about numerous control paths resulting from the complex interleaving of events (messages) and failures.
Unsurprisingly, it is easy to introduce subtle errors while attempting to fill... More >
Seminar: Dissertation Talk: EE | May 3 | 3-4 p.m. | 250 Sutardja Dai Hall
As robots become more capable and commonplace, it is increasingly important that the policies they execute are transparent. For instance, engineers should have an idea of which situations their robot may act incorrectly in, and end-users should be able to anticipate how a robot they are interacting with will behave in various situations. This is essential for building trust, enabling seamless... More >
Presentation: Dissertation Talk: CS | May 6 | 9:30-10:30 a.m. | 380 Soda Hall
This thesis describes fast and accurate RTL simulation methodologies for performance, power, and energy evaluation as well as verification and debugging using FPGAs in the hardware/software co-design flow.
Seminar: Dissertation Talk: CS | May 6 | 1-2 p.m. | 405 Soda Hall
Due to the prevalence of machine learning (ML) algorithms and the potential for their decisions to profoundly impact billions of human lives, it is crucial that they are robust, reliable, and understandable. This thesis examines key theoretical pillars of ML surrounding generalization and overfitting, and tests the extent to which empirical behavior matches existing theory. We develop novel... More >
Seminar: Dissertation Talk: CS | May 6 | 2-3 p.m. | 606 Soda Hall
Secure computation enables a set of mutually distrusting parties to collaboratively compute a public function over their private data while keeping those data private. A central question in the cryptography community, that has been studied for decades, is whether secure computation protocols can be efficient enough to serve for its countless applications. In this talk, I will present my efforts... More >
Seminar: EE: Berkeley Laboratory for Information and System Sciences (BLISS) | May 6 | 3-4 p.m. | 540 Cory Hall
Anand Sarwate, Rutgers
The information theory community has traditionally studied two different models for communication. The Shannon-theoretic model treats the channels impact as random, so codes must correct almost all error patterns of a given weight; this is an average-case analysis. The coding-theoretic (Hamming-theoretic?) model treats the channel as adversarial, so codes must correct all error patterns of a... More >
Conference/Symposium: EE | May 6 | 4:30-8:30 p.m. | Sibley Auditorium, Bechtel Engineering Center
Come expand your network at Space Tech Symposium 2.0 @ Berkeley (https://stac.berkeley.edu/sts2) on May 6 by meeting researchers, CEOs of the hottest space startups, and Berkeley faculty as they discuss their visions for the future of space development.
Mobility between space and non-space fields is at an all-time high and we'd love to have you join this conversation. Panelists from NASA,... More >
$10 Student Tickets, $30 Industry / Faculty Tickets
Presentation: Dissertation Talk: CS | May 7 | 1-2 p.m. | 521 Cory Hall
The advent of algorithms capable of leveraging vast quantities of data and computational resources has led to the proliferation of systems and tools aimed to facilitate the development and usage of these algorithms. Hardware trends, including the end of Moore's Law and the maturation of cloud computing, have placed a premium on the development of scalable algorithms designed for parallel... More >
Seminar: Dissertation Talk: CS | May 7 | 2-3 p.m. | 310 Soda Hall
Pasin Manurangsi, University of California, Berkeley
The theory of NP-hardness of approximation has led to numerous tight characterizations of approximability of hard combinatorial optimization problems. Nonetheless, there are many fundamental problems which are out of reach for these techniques, such as problems that can be solved (or approximated) in quasi-polynomial time... More >
Seminar: Dissertation Talk: EE | May 7 | 4-5 p.m. | 490 Cory Hall
In this talk I will discuss my work on a crystal-free IEEE 802.15.4 receiver for the Single Chip Mote project.
Seminar: Dissertation Talk: EE | May 8 | 12-1 p.m. | 337A Cory Hall
In recent times, the field of computer vision has made great progress with recognizing and tracking people and their activities in videos. However, for systems designed to interact dynamically with humans, tracking and recognition are insufficient; the ability to predict behavior is requisite. In this talk, I will present my work on learning to make predictions from visual input. Using team... More >
Seminar: Dissertation Talk: EE | May 8 | 12:30-1:30 p.m. | Soda Hall, Visual Computing Lab (VCL)
Michael Chen, UC Berkeley
Imaging biological samples under optical microscopes is challenging, since the absorption is too weak to form images with informative contrast. Besides fluorescent imaging techniques, label-free phase contrast imaging methods have been proposed to greatly improve the contrast of transparent samples. In order to efficiently recover quantitative properties, such as 2D phase projection and 3D... More >
Scientific Computing and Matrix Computations Seminar: A Backward Error Formula for the Global Least-Squares Problem
Seminar: Scientific Computing: CS | May 8 | 2-3 p.m. | 380 Soda Hall
Eric Hallman, UCB
In this talk we present formulas for the normwise backward error for the problems AX=B and min_X ||AX-B||_F, extending the results of Walden/Karlson/Sun (1998) and Sun/Sun (1997) to systems with multiple right hand sides. These formulas are valid for matrices of any dimension and rank. We also propose two estimates that numerical experiments suggest are highly reliable, one of which may be... More >
Seminar: Dissertation Talk: EE | May 8 | 4-5 p.m. | Cory Hall, 490H (Immersion Room in Swarm Lab)
Taehwan Kim, UC Berkeley
LiDAR (Light Detection and Ranging) captures high-definition real-time 3D images of the surrounding environment, which makes it a crucial sensing modality for applications such as self-driving cars. However, high price tag of existing commercial LiDAR modules based on mechanical beam scanners and intensity-based detection scheme prohibits them from being extensively applied to consumer products.... More >
Lecture: Dissertation Talk: CS | May 8 | 4-5 p.m. | 540AB Cory Hall
While state-of-the-art machine learning models are deep, large-scale, sequential and highly nonconvex, the backbone of modern learning algorithms are simple algorithms such as stochastic gradient descent, or Q-learning (in the case of reinforcement learning tasks). A basic question endures---why do simple algorithms work so well even in these challenging settings?
This talk focuses on two... More >
Presentation: Dissertation Talk: EE | May 9 | 9-10 a.m. | 540AB Cory Hall
Li-Hao Yeh, UC Berkeley
High-content microscopy targets high-resolution imaging across large fields-of-view (FOVs). Recent works have demonstrated that computational imaging, which generalizes the information... More >
Seminar: Dissertation Talk: CS | May 9 | 10-11 a.m. | 465H Soda Hall
Anurag Khandelwal, University of California at Berkeley
Modern cloud data services aim to support increasingly sophisticated queries with interactive response times. These services can be broadly divided into two categories: read-intensive applications such as web services, and write-intensive applications such as real-time monitoring of event streams. In both cases, supporting sophisticated queries interactively and at scale raises significant... More >
Colloquium: Dissertation Talk: CS | May 9 | 11 a.m.-12 p.m. | 606 Soda Hall
Presentation: Dissertation Talk: EE | May 9 | 11 a.m.-12 p.m. | 531 Cory Hall
It is often easier to describe what we want an autonomous system to do rather than how to do it. A control synthesizer bridges that gap by automatically constructing control software so the closed loop dynamics enforce a desired behavior such as safety or reachability. While many specific instances of control synthesis have elegant mathematical solutions, designing tractable algorithms to compute... More >
Presentation: Dissertation Talk: EE | May 9 | 1-2 p.m. | Cory Hall, 521 Hogan room
Palak Bhushan, EECS Dept, UC Berkeley
In this talk we will look at the design of 0.1gram centimeter-scale microbots utilizing different locomotion strategies as a means of transport. The focus won't be on the brain or useful payloads for these bots, but instead on the electro-mechanical design to make these tether-less, which is a necessary precursor to making autonomous microbots.
We start with the design of a micro-ratcheting... More >
Seminar: Dissertation Talk: EE | May 9 | 3:30-4:30 p.m. | Cory Hall, 490H (Immersion Room in Swarm Lab)
Efthymios Philip Papageorgiou
This talks presents a chip-scale fluorescence microscope for the detection of microscopic residual disease, small clusters of hundreds to thousands of cancer cells left behind after the gross tumor is removed during a surgical resection.
Seminar: Dissertation Talk: CS | May 10 | 9-10 a.m. | 606 Soda Hall
Machine learning is being deployed in a growing number of applications which demand real-time, accurate, and cost-efficient predictions under heavy query load. These applications employ a variety of machine learning frameworks and models, often composing several models within the same application. However, most machine learning frameworks and systems are optimized for model training and not... More >
Special Event: Departmental: EE: CS | May 10 | 11 a.m.-1:30 p.m. | Sibley Auditorium, Bechtel Engineering Center
Each year the EECS Student Awards Committee selects winners for each of our department awards, many based on nominations gathered from EECS students, faculty and staff.
Dissertation Talk: Expert-Level Detection of Acute Intracranial Hemorrhage on Head Computed Tomography using Deep Learning
Presentation: Dissertation Talk: EE | May 10 | 1-2 p.m. | Sutardja Dai Hall
Head CT is the workhorse medical imaging modality used worldwide to diagnose neurologic emergencies. We demonstrated state-of-the-art exam-level classification performance, comparable to that of U.S. board-certified radiologists, in addition to robust localization of abnormalities, both of which are important to this application.
Presentation: Dissertation Talk: CS | May 10 | 1-2 p.m. | 380 Soda Hall
In the talk we will present our work in designing components for an AutoML solution.
Solid State Technology and Devices Seminar: High-Q chalcogenide device platform without direct etching process for non-linear and mid-IR Applications
Seminar: Solid State Technology and Devices: EE: CS | May 10 | 1-2 p.m. | Cory Hall, The Hogan Room, 521
Sangyoon Han, Postdoc, KAIST, Korea
In this talk, I will show our new device platform that defines chalcogenide devices without direct etching process. Using our platform, we have demonstrated chalcogenide resonators with record high (for on-chip chalcogenide devices) Q-factor (1.2 x 10^7) and Brillouin lasers with record low (for on-chip chalcogenide devices) threshold power (2.5mW).
Seminar: Dissertation Talk: EE | May 10 | 2-3 p.m. | 250 Sutardja Dai Hall
Yoonhwan Kang, UC Berkeley
Lowering power consumption of electronics is driven by three reasons: i) the worldwide energy consumption of semiconductor chips is a sizable contributor to the global warming, ii) the high-power density of the chips is one of the main obstacles to the scalability of the transistors, and iii) the market of Internet-of-Things has risen to a nontrivial sector of the global electronics market.... More >
Seminar: Dissertation Talk: EE | May 10 | 3-4 p.m. | Cory Hall, 540A/B
Tianshi Wang, UC Berkeley
Over the last few years, there has been considerable interest in Ising machines, ie, analog hardware for solving difficult (NP hard/complete) computational problems effectively. We present a new way to make Ising machines using networks of coupled self-sustaining nonlinear oscillators. Our scheme is theoretically rooted in a novel result that connects the phase dynamics of coupled oscillator... More >
Seminar: Dissertation Talk: EE | May 10 | 3-4 p.m. | 540 Cory Hall
I will show that the phase dynamics of sub-harmonically injection locked coupled oscillator systems are governed by a Lyapunov function that continuizes the Ising Hamiltonian.
This result is used to devise a new Ising machine scheme that can use oscillators from many different physical domains.
A demonstration of a prototype CMOS oscillator-based Ising machine of 240 spins will follow the talk.
Seminar: Dissertation Talk: CS | May 10 | 3:30-4:30 p.m. | 373 Soda Hall
The problem of generating a large number of diverse solutions to a logical constraint has important applications in testing, verification, and synthesis for both software and hardware. The solutions generated could be used as inputs that exercise some target functionality in a program or as random stimuli to a hardware module. The sampling of solutions can be combined with techniques such as fuzz... More >
Lecture: Dissertation Talk: CS | May 13 | 10-11 a.m. | 405 Soda Hall
This dissertation talk explores how to improve both the utility of shared data and how well privacy is maintained with strategically designed tools and methods. We propose and evaluate these tools and strategies for collaborative data management to help navigate tensions between open data and data privacy in the context of international development engineering projects.
Lecture: Dissertation Talk: CS | May 13 | 11 a.m.-12 p.m. | 405 Soda Hall
Shaddi Hasan, UC Berkeley
Over 400 million people live without access to basic communication services, largely in rural areas. Community-based networks, and particularly community cellular networks, can sustainably support services even in these extremely rural areas where commercial network operators cannot.
In this talk, I will identify key challenges these community cellular networks face for reaching scale, and... More >
Seminar: Dissertation Talk: EE | May 13 | 1:30-2:30 p.m. | 531 Cory Hall
Shromona Ghosh, University of California, Berkeley
We are finally at a point where we have the knowledge and resources to make safety-critical robotic systems a reality. Deploying such systems in the real world, however, requires addressing problems in detecting unsafe environments, reasoning with unknown or learnt components and providing strong safety assurances. In this talk, I will address two specific challenges in this domain: (1)... More >
Seminar: Dissertation Talk: CS | May 13 | 2-3 p.m. | 380 Soda Hall
Philipp Moritz, UC Berkeley
In recent years, growing data volumes and more sophisticated computational procedures have greatly increased the demand for computational power. Machine learning and artificial intelligence applications, for example, are notorious for their computational requirements. At the same time, Moores law is ending and processor speeds are stalling. As a result, distributed computing and the cloud have... More >
Seminar: Dissertation Talk: EE | May 13 | 3-4 p.m. | Cory Hall, 540A/B
Silicon photonics is a rapidly maturing platform for optical communication and sensing. As systems leveraging silicon photonics have grown in size and complexity, so too has the demand for high performance silicon photonics components. In order to meet these demands, we propose a hierarchical approach to design and optimization of silicon photonics components. Our approach applies simple physical... More >
Seminar: Dissertation Talk: EE | May 14 | 2:30-3:30 p.m. | Cory Hall, 400 Hughes Room
Jonas Kapraun, UC Berkeley
Vertical cavity surface emitting lasers (VCSELs) have been widely employed in short distance optical interconnects. Recently however a series of emerging applications are creating a rapidly growing demand for compact, low cost and high-performance light sources... More >
Dissertation Talk: Monolayer Transition Metal Dichalcogenide NanoLEDs: Towards High Speed and High Efficiency
Seminar: Dissertation Talk: EE | May 14 | 2:30-4 p.m. | 540AB Cory Hall
Monolayer transition metal dichalcogenides (TMDCs) are a class of 2D semiconductors with attractive properties for future nanoscale optoelectronics such as light sources and detectors. However, due to their single molecule thickness, they are sensitive to environmental factors such as moisture, limiting operation of most light-emitting devices to high vacuum conditions. In this talk, I will... More >
Seminar: Dissertation Talk: EE | May 15 | 9-10 a.m. | 521 Cory Hall
In recent decades, we have witnessed the emergence of an electric grid that is increasingly decentralized and is capable of reliably integrating renewable resources that are distributed and intermittent in nature. While this important trend continues with the electric grid, there is growing need for a similar decentralization of electric networks in a variety of high-impact emerging applications,... More >
Lecture: Dissertation Talk: CS | May 15 | 10-11 a.m. | 360 Hearst Memorial Mining Building
Cesar Torres, University of California Berkeley
Creative technologies like digital fabrication led to the rise of the Maker Movement, engendering grassroots innovation in education, manufacturing, and healthcare. This talk lays down a framework for composing new materials and technologies to foreground the existing knowledge and practices of material practitioners and alter the trajectory of the Maker Movement towards a New Making Renaissance.
Presentation: Dissertation Talk: CS | May 15 | 10-11:10 a.m. | 310 Soda Hall
Yi Wu, UC Berkeley
Lecture: Dissertation Talk: CS | May 15 | 11 a.m.-12 p.m. | Soda Hall, 306 (HP Auditorium)
Ke Li, Ph.D. Candidate, UC Berkeley
In this talk, I will present our work on overcoming two long-standing problems in machine learning and algorithms: the curse of dimensionality in nearest neighbour search and mode collapse in generative adversarial nets (GANs).
Seminar: Dissertation Talk: CS | May 15 | 1:30-2:30 p.m. | Sutardja Dai Hall, Newton Room/730
Evan Shelhamer, UC Berkeley
Much of the recent progress on visual recognition has been driven by deep learning and its bicameral heart of composition and end-to-end optimization. Its diffusion however was neither instantaneous nor effortless.... More >
Presentation: Dissertation Talk: CS | May 15 | 2-3 p.m. | 405 Soda Hall
Seminar: Dissertation Talk: EE: CS | May 15 | 3-4 p.m. | 540AB Cory Hall
Matt Weber, UC Berkeley
Future IoT killer apps leveraging ubiquitous sensors and actuators will create value in emergent properties of composition and contextual awareness. This talk focuses on two key aspects of principled IoT design: (1) using contextual information from the physical world, and (2) enabling interaction and composition across distributed cyber physical systems. These challenges are mutually... More >
Presentation: Dissertation Talk: EE | May 16 | 2108 Allston Way (Berkeley Wireless Research Center), Rabaey Room
A frequency-flexible radio-frequency (RF) front end has long been desired, but faces a myriad of obstacles to its realization. In recent years, the use of switching power amplifiers (PA) as part of digital PAs and RF Digital-to-Analog Convers (RFDAC) has become more common. The primary motivation of these RFDACs is to directly convert from digital baseband bits to RF output. This is useful in... More >
Presentation: Dissertation Talk: EE | May 16 | 8-9 a.m. | 2108 Allston Way (Berkeley Wireless Research Center), Rabaey Room
Various receive side SerDes techniques including an integrating CTLE, FFE and DFE are demonstrated in a 60Gb/s NRZ transceiver consuming 288 mW & occupying 2.48mm2. To support burst mode, a 2-tap SC FFE transmitter w/ a rapid-ON/OFF VCO is presented that allows for the lowest achievable latency for a 64:1 1-latch SER, achieving 1.2ns startup time and 0.72-0.62 pJ/bit at 1-20Gb/s occupying 0.19mm2.
Presentation: Dissertation Talk: EE | May 16 | 9-10 a.m. | 405 Soda Hall
Anand Padmanabha Iyer
Dissertation talk: Global Data Plane -- An architecture for a globally distributed storage and communication infrastructure
Presentation: Dissertation Talk: CS | May 16 | 11 a.m.-12 p.m. | 606 Soda Hall
In today's world, the storage and management of information is highly centralized in data-centers. Such a data-center oriented view of infrastructure also shapes the way we think about information security; the de-facto mechanism for security is to draw a perimeter around the infrastructure and protect such perimeter. However, the future is very different. With the advancement of technology,... More >
Presentation: Dissertation Talk: CS | May 16 | 3:30-4:30 p.m. | 2108 Allston Way (Berkeley Wireless Research Center)
James C. Martin
In recent years, there has been a trend of increasing capacity and carrier frequency in wireless networks design. This talk will cover some of the insights that happen when constructing large mesh networks using directional antenna arrays at mmWave.
Presentation: Dissertation Talk: CS | May 17 | 12-1 p.m. | Soda Hall, 510 (VCL)
Deepak Pathak, Computer Science, UC Berkeley
Generalization, i.e., the ability to adapt to novel scenarios, is the hallmark of human intelligence. While we have systems that excel at recognizing objects, cleaning floors, playing complex games and occasionally beating humans, they are incredibly specific in that they only perform the tasks they are trained for and are miserable at generalization. Could actually optimizing towards... More >
Miscellaneous: Departmental: EE | May 21 | 9-11:30 a.m. | Hearst Greek Theatre
This ceremony includes all departments in the College of Engineering.