RSS FeedUpcoming EventsDissertation Talk: A Data Converter Assisted Beamforming Method, March 20https://events.berkeley.edu/eecs/event/243522-dissertation-talk-a-data-converter-assisted

Directional communication, a key feature in 5G technology and beyond, expands channel capacity by leveraging spatial information from a multiple-antenna front-end. Fully digital beamforming enables powerful signal processing techniques, such as spatial blocker suppression and beam search. However, in the presence of spatial blockers, the digital back-end requires high dynamic range data converters to preserve information. This has led to increased interest in research on spatial filtering or beamforming in the analog/RF front-end. While inserting a spatial filter in the signal path can attenuate spatial blockers at an early stage, it usually results in an increased power budget and additional noise. In this work, we propose a data converter-assisted beamforming method, which leverages the computational capability of the data converters and utilizes existing circuitry. The proposed method has low hardware and power overheads and is compatible with other spatial filtering methods.

https://events.berkeley.edu/eecs/event/243522-dissertation-talk-a-data-converter-assisted
Communicating Algorithmic Science to the Public, March 20/live/events/243102-communicating-algorithmic-science-to-the-public

Algorithms increasingly pervade every aspect of daily life. The importance of this societal development is widely acknowledged, but how much does the public understand about the underlying science? This panel discussion brings together a theoretical computer scientist with science communicators specializing in math and computer science, to explore the question of how to communicate algorithmic science to a broad audience.

Alex Bellos is a British popular-science writer whose books have sold more than a million copies and been translated into more than 20 languages. He is the puzzle columnist for The Guardian, as well as one of the main presenters of the YouTube channel Numberphile, and has presented science documentaries for BBC Radio. He is currently writing a book about theoretical computer science for the general reader, which will be published by Knopf in the United States.

Ananyo Bhattacharya is chief science writer at the London Institute for Mathematical Sciences. During a 15-year career in journalism, he has worked as a senior editor at Nature, Chemistry World, and Research Fortnight, and as a community editor and science correspondent for The Economist. He is the author of The Man from the Future, an intellectual biography of John von Neumann.

Ben Brubaker is a New York City–based science journalist who covers theoretical computer science for Quanta Magazine. Before joining Quanta as a staff writer, he covered physics as a freelancer for publications including Scientific American and Physics Today. He holds a PhD in physics from Yale and conducted postdoctoral research at the University of Colorado, Boulder.

Sampath Kannan is the Henry Salvatori Professor in the Department of Computer and Information Science at the University of Pennsylvania. His research interests are in the areas of algorithmic fairness, combinatorial algorithms, program reliability, streaming computation, and computational biology. He is a fellow of the ACM, and the recipient of the ACM SIGACT Distinguished Service Award. He is also a fellow of the American Association for the Advancement of Science.


Theoretically Speaking is a lecture series highlighting exciting advances in theoretical computer science for a broad general audience. Events are free and open to the public, with first-come, first-served seating. No special background is assumed. Registration is required. This lecture will be viewable on our YouTube channel following captioning.

Light refreshments will be provided before the talk, starting at 5:15 p.m.

The Simons Institute regularly captures photos and video of activity around the Institute for use in publications and promotional materials.

/live/events/243102-communicating-algorithmic-science-to-the-public
New Silicon Initiative: Women’s History Month Special Event, March 21https://events.berkeley.edu/eecs/event/243361-new-silicon-initiative-womens-history-month

YOU MUST REGISTER HERE
(Zoom details will be shared with you upon registering)

Celebrate Women’s History Month with Anuja Banerjee, Silicon Engineering Group, and Heather Sullens, Exploratory Design Group — senior leaders from Apple’s hardware technologies.

Anuja and Heather will share their career journeys and offer insightful perspectives on what it’s like to lead teams at the forefront of solving some of the most complex engineering challenges.

Interested in opportunities at Apple?

Let us know by submitting your resume and application at Careers at Apple.

 

Register Here

https://events.berkeley.edu/eecs/event/243361-new-silicon-initiative-womens-history-month
International House Celebration & Awards Gala, March 21/live/events/239141-international-house-celebration-awards-gala

International House (I-House) at UC Berkeley will hold its Annual Celebration and Awards Gala on Thursday, March 21, 2024. I-House proudly celebrates and honors advocates and trailblazers who embody the purpose of I-House to promote a more just and peaceful world. This year we are pleased to highlight the contributions of the following individuals:

Chenming Hu (IH 1969-71) recipient of the Global Impact Award for his contributions to the semiconductor industry which have led to transformative improvements in computing and communications around the globe;

Chiara Medioli-Fedrigoni (IH 1993-94) recipient of the Alumna of the Year Award for promoting cultural heritage preservation efforts at universities, museums, archives, and libraries in Europe and throughout the world;

Okechukwu Iroegbu (IH 2022-24) recipient of the Executive Director’s Outstanding Community Leadership Award for his dedication to the mission of I-House and the positive impact his leadership has had on the resident community;

Ronald E. Silva (I-House Board Member) recipient of The Sherry and Betsey Warrick Mission Service Award for his dedication to the preservation and improvement of I-House.

Read their bios at ihouse.berkeley.edu/gala.

Proceeds from the Gala benefit The Fund for I-House. Shaun R. Carver, Executive Director I-House, points to this Fund as an important resource, providing for room and board scholarships and financial aid, mission-centered programming, and preservation of the historic 93-year-old building. “I-House, a 501(c)3 nonprofit organization, relies on philanthropic support from alumni, foundations, corporations and friends,” says Carver.

Reception and dinner will be provided by award-winning Executive Chef Abigail Serbins and our Dining, Catering and Events teams. Entertainment will be provided by the very talented residents of International House. For more information on our honorees, sponsorship opportunities, or to buy event tickets, visit ihouse.berkeley.edu/gala

/live/events/239141-international-house-celebration-awards-gala
Designing Advanced Nanocatalysts by Looking at Atoms and Molecules on Reactive Surfaces: Nano Seminar series, March 22/live/events/237676-designing-advanced-nanocatalysts-by-looking-at

Clarification of the nature of active sites at both solid-gas and solid-liquid interfaces has been a long-standing question in surface chemistry, holding paramount significance in crafting innovative catalytic materials that demand minimal energy consumption. A bimetallic Pt alloy, or mixed catalyst, is an excellent platform to uncover the contentious role of the metal–metal oxide interface because the alloyed transition metal can coexist with the Pt surface layer in the form of an oxidized species on the bimetal surface during catalytic reactions. The real-time imaging of catalytically reactive atomic sites using operando surface techniques, including ambient pressure scanning tunneling microscopy, can reveal the nature of reactive sites on the catalytic surfaces.

In this talk, I present in-situ observation results of structural modulation on Pt-based bimetal catalysts and mixed catalysts and its impact on the catalytic activity. We utilized PtNi, and PtCo that includes both single crystal and nanoparticle surfaces as model catalysts, and showed the coexistence of Pt and metal oxide leads to the enhancement of catalytic activity, indicating these metal-oxide interfaces provide a more-efficient reaction path for CO oxidation. The mixed catalysts composed of Pt nanoparticles and the mesoporous cobalt oxide exhibit the enhancement of catalytic activity while Pt is encapsulated by the oxide thin layers forming the reactive metal-oxide interfaces. In addition, we address the fundamentals of the electrocatalytic process and on locating the real active sites at the solid-liquid interface by utilizing in-situ electrochemical scanning tunneling microscopy. Overall, the atomic-scale imaging of the reactive surfaces gives rise to the design rule of advanced bimetallic and mixed catalysts.

**************

Jeong Young Park (박정영) did his PhD in Physics at Seoul National University and postdoc at LBNL (Go Bears!) After some years as staff scientist here he returned to Korea and joined the Chemistry faculty at KAIST. Prof. Park has authored 320 peer-reviewed papers and book chapters in international journals.

 

/live/events/237676-designing-advanced-nanocatalysts-by-looking-at
CLIMB Evergreen talk with Pradeep Ravikumar: The Return of the Kernel Method: Augmentation-Based Self-Supervised Representation Learning & RKHS regression, April 5/live/events/239709-climb-evergreen-talk-with-pradeep-ravikumar

The Return of the Kernel Method: Augmentation-Based Self-Supervised Representation Learning & RKHS regression

Abstract: 

Modern self-supervised representation learning approaches such as contrastive learning and masked language modeling have had considerable empirical successes. A typical approach to evaluate such representations involves learning a linear probe on top of such representations and measuring prediction performance with respect to some downstream prediction task. We add to a burgeoning understanding of such representation learning approaches as learning eigenfunctions of certain Laplacian operators, so that learning a linear probe can be naturally connected to RKHS regression with implicitly specified kernels. This allows us to extend non-parametric tools from RKHS regression analysis to analyze the performance of self-supervised representation learning methods, in a way that completely side-steps grappling with neural network based function classes used in practice for the representation encoders. This contravenes prevailing wisdom that we cannot understand these modern representation learning methods without understanding the inductive bias implicit in the intersection of deep neural networks and the optimization methods used to learn them. We specifically focus on augmentation based self-supervised learning approaches, and isolate key structural complexity characterizations of augmentations that we show can be used to quantitatively inform downstream performance of the learned representations.

Bio: 

Pradeep Ravikumar is a Professor in the Machine Learning Department, School of Computer Science at Carnegie Mellon University. His thesis has received honorable mentions in the ACM SIGKDD Dissertation award and the CMU School of Computer Science Distinguished Dissertation award. He is a Sloan Fellow, a Siebel Scholar, a recipient of the NSF CAREER Award, and co-editor-chief of the Journal of Machine Learning Research. His recent research interests are in neuro-symbolic AI, combining statistical machine learning, and symbolic and causal learning.
/live/events/239709-climb-evergreen-talk-with-pradeep-ravikumar
Polar Topological Defects – Fundamentals to Applications: Nano Seminar series, April 5/live/events/230393-polar-topological-defects-fundamentals-to

Topological defects such as vortices and skyrmions have recently gained significant interest in solid state materials as ferroic materials (ferromagnets and ferroelectrics) have become a test-bed to realize and control these nanoscale structures. Although this phenomenon is being investigated as a pathway to energy efficient information storage, broader applications in interaction of electromagnetic waves with such features are emerging.

In the case of ferroelectrics, boundary condition engineering is used to achieve vortices, skyrmions, and merons in low dimensional epitaxial oxide heterostructures.

In this talk, I will introduce the notion that similar phenomenology but at the atomic scale can be achieved in charge density wave phases, especially nominally semiconducting chalcogenides. I will outline my group and other groups’ efforts in showing non-trivial toroidal polar topologies at the atomic level in chalcogenides with nominally empty conduction band with d-orbital character such as 1T-TiSe2, Ta2NiSe5 and BaTiS3.

Specifically, we use X-ray single crystal diffraction as a probe for high quality single crystals of a quasi-1D hexagonal chalcogenide, BaTiS3, to reveal complex polar topologies such as vortices, and head-to-head and tail-to-tail arrangement of dipoles. Recent experiments and theoretical studies on the stability and dynamics of these features, and their broad connection to low dimensional magnets, will also be discussed. Lastly, I will outline the perspective for photonic applications of polarization textures.

***********

Jayakanth Ravichandran did his PhD in the Ramesh lab here at UC Berkeley (Go Bears!) and postdoc with Philip Kim at Harvard. He joined the USC faculty in 2015.

/live/events/230393-polar-topological-defects-fundamentals-to
Dissertation Talk: Object-Centric Perception for Real-World Robotics, April 12https://events.berkeley.edu/eecs/event/238105-dissertation-talk-object-centric-perception-for

Recent advances in deep learning have resulted in remarkable progress in many areas. However, these methods are not sufficient for many robotics applications, due to the long-tail visual diversity of the real world, stringent requirements for autonomous, high-throughput operation, and the lack of large-scale training datasets. In this talk, I will discuss how we address these challenges via improved methods for 2D and 3D perception. First, I will introduce techniques that enable perception models to better express uncertainty in challenging or ambiguous situations. Robots equipped with these models can more explicitly reason about the inherent ambiguity of the real world in the context of their particular tasks. Next, I will discuss how we can leverage object-centric NeRFs to more easily obtain real-world supervision for such perception models. I will show how these methods can be used to generate training data targeted to specific real-world environments, for a variety of perceptual tasks.

https://events.berkeley.edu/eecs/event/238105-dissertation-talk-object-centric-perception-for
CLIMB Evergreen Talk with Martin Wainwright, April 12/live/events/243186-climb-evergreen-talk-with-martin-wainwright

CLIMB Evergreen talk with Martin Wainwright. More information coming

/live/events/243186-climb-evergreen-talk-with-martin-wainwright
Dissertation Talk: Scalable Lifelong Imitation Learning for Robot Fleets, April 17https://events.berkeley.edu/eecs/event/237987-dissertation-talk-scalable-lifelong-imitation

Recent breakthroughs in deep learning have revolutionized natural language processing, computer vision, and robotics. Nevertheless, reliable robot autonomy in unstructured environments remains elusive. Without the Internet-scale data available for language and vision, robotics faces a unique chicken-and-egg problem: robot learning requires large datasets from deployment at scale, but robot learning is not yet reliable enough for deployment at scale. We propose a scalable human-in-the-loop learning paradigm called “interactive fleet learning” (IFL) as a potential solution to this paradox, and we argue that it is the key ingredient behind the recent growth of large-scale robot deployments in applications such as autonomous driving and e-commerce order fulfillment. We develop novel formalisms, algorithms, benchmarks, systems, and applications for IFL and evaluate its performance in extensive simulation and real world experiments. We conclude with a discussion of limitations and opportunities for future work.

https://events.berkeley.edu/eecs/event/237987-dissertation-talk-scalable-lifelong-imitation
Dissertation Talk: The management of data context in the machine learning lifecycle, April 19https://events.berkeley.edu/eecs/event/239965-dissertation-talk-the-management-of-data-context-in-t

In this dissertation talk, I present groundbreaking research on managing data context—–Application, Behavior, and Change over time—within the machine learning (ML) lifecycle. Drawing from a vision laid out in a 2018 KDD workshop, my work introduces Flor (VLDB ’21) and its evolutions, FlorDB (VLDB ’24) and FlorFlow (ArXiv pre-print), designed for comprehensive metadata capture, version control, and provenance analysis in ML model management. A cornerstone of our approach is the use of an interview study to understand “How engineers operationalize machine learning” (CSCW ’24), focusing on MLOps and the iterative model development process.

Through the implementation of these systems and their use in real-world applications for law and journalism, my research demonstrates the tangible benefits of rich data context in agile model development. This talk will demonstrate how the integration of Application, Behavior, and Change contexts enhances the management of the ML lifecycle.

https://events.berkeley.edu/eecs/event/239965-dissertation-talk-the-management-of-data-context-in-t
Dissertation Talk: Solving Matrix Sensing to Optimality under Realistic Settings, April 24https://events.berkeley.edu/eecs/event/243595-dissertation-talk-solving-matrix-sensing-to

Matrix sensing represents a critical, non-convex challenge within the domain of mathematical optimization, distinguished by its wide-ranging practical applications—such as medical imaging, recommender systems, and phase retrieval—as well as its significant theoretical contributions, particularly its equivalence to training a two-layer quadratic neural network. The ability to efficiently solve this problem to optimality promises substantial benefits not only for its direct applications but also provides a crucial benchmark that aids in navigating the increasingly intricate non-convex landscapes characteristic of contemporary machine learning systems. While prior research predominantly focuses on scenarios abundant in observations and characterized by a low Restricted Isometry Property (RIP) constant, thereby facilitating optimal solutions through either convex relaxation methods, including nuclear-norm minimization, or local search strategies applied to the Burer-Monteiro factorized formulation—thereby accelerating computational processes without compromising performance guarantees—the research to date remains incomplete. This is particularly true in real-world settings where acquiring a large volume of observations is often impractical, thus rendering these guarantees inapplicable.

In this dissertation, we propose innovative strategies, models, and conceptual frameworks aimed at addressing the matrix sensing problem under conditions of limited observations and noise corruptions, with the objective of provably reconstructing the ground truth matrix. Our discussion begins by exploring various methodologies of over-parametrization as a means to solve this problem, followed by an examination of alternative solutions in scenarios where over-parametrization is not used. Additionally, we delve into the impact of noise on the extraction of a global solution, offering insights into how it affects the overall process. This work serves not only as an elaborate guide to resolving matrix sensing and, by extension, low-rank optimization problems in less than ideal conditions but also endeavors to enhance our understanding of the complexities involved in non-convex optimization, thereby contributing to the broader field of mathematical optimization and machine learning.

https://events.berkeley.edu/eecs/event/243595-dissertation-talk-solving-matrix-sensing-to
Dissertation Talk: Connect the Dots: Modeling Social Interactions from Multimodal Signals, May 6https://events.berkeley.edu/eecs/event/242459-dissertation-talk-connect-the-dots-modeling-social

As social agents, humans have social intelligence that enables them to engage with others through complex signals such as facial expressions, body motion, and speech, among other communication modalities. Building systems that can perceive and model these fine-grain interaction dynamics is therefore essential for advancing human-machine interactions in everyday life. In this talk, I will share three of our recent advancements in this research area. The first one explores modeling nonverbal dyadic conversational dynamics between the facial expressions of a speaker and listener. The second one builds upon this prior work, expanding to full-body dynamics. Our final work then explores how we can incorporate higher-level syntactic understanding by leveraging large language models.

https://events.berkeley.edu/eecs/event/242459-dissertation-talk-connect-the-dots-modeling-social
Dissertation Talk: Bridging Gaps Between Metrics and Goals in Modern Machine Learning Ecosystems, May 8https://events.berkeley.edu/eecs/event/241810-dissertation-talk-bridging-gaps-between-metrics

A prevailing challenge in understanding and improving the societal impacts of machine learning (ML) systems stems from the fact that numerical metrics used for optimization, auditing, or accountability do not always match broader social goals. For example, higher predictive accuracy of an ML model for predicting student dropout risk does not always yield better student outcomes downstream. Such mismatches between metrics and goals present a challenge to both policymakers seeking to audit algorithmic systems, and engineers and researchers formulating ML problems.

In this talk, I will discuss theoretical, algorithmic, and qualitative approaches to bridging gaps between metrics and goals. First, I will discuss recent work on understanding how metrics fit into an ecosystem of stakeholders – specifically, I will show how causal metrics can improve social welfare in ranking systems when entities being ranked can strategically respond. Second, I will discuss implementation gaps between theory and practice in Fair ML, using robust optimization approaches to handle noisy data. Finally, I will touch on a qualitative interview study in ML applied in education settings, and discuss open questions towards better problem formulations when ML fits into social contexts with interactions between stakeholders.

https://events.berkeley.edu/eecs/event/241810-dissertation-talk-bridging-gaps-between-metrics
Class of 2024 Engineering Master’s Degree Commencement, May 14/live/events/229317-class-of-2024-engineering-masters-degree

The College of Engineering will host a commencement ceremony for Master’s degree graduates of the Class of 2024, their family and friends on Tuesday, May 14.

/live/events/229317-class-of-2024-engineering-masters-degree
Class of 2024 Engineering Baccalaureate Degree Commencement, May 14/live/events/242509-class-of-2024-engineering-baccalaureate-degree-commen

The College of Engineering will host a commencement ceremony for Baccalaureate degree graduates of the Class of 2024, their family and friends on Tuesday, May 14.

/live/events/242509-class-of-2024-engineering-baccalaureate-degree-commen
Class of 2024 Engineering Doctoral Degree Commencement, May 18/live/events/242512-class-of-2024-engineering-doctoral-degree-commencemen

The College of Engineering will host a commencement ceremony for Doctoral degree graduates of the Class of 2024, their family and friends on Saturday, May 18.

/live/events/242512-class-of-2024-engineering-doctoral-degree-commencemen