EECS alumnus Yang You (Ph.D. '20, advisor: James Demmel) was named as one of two honorable mentions for the 2020 ACM Special Interest Group in High Performance Computing (SIGHPC) Dissertation Award. You was selected for developing LARS (Layer-wise Adaptive Rate Scaling) and LAMB (Layer-wise Adaptive Moments for Batch training) to accelerate machine learning on HPC platforms. His thesis, “Fast and Accurate Machine Learning on Distributed Systems and Supercomputers,” focuses on improving the speed and accuracy of Machine Learning training to optimize the use of parallel programming on supercomputers. You made the Forbes 30 Under 30 2021 Asia list for Healthcare and Science in April and is now a Presidential Young Professor of Computer Science at the National University of Singapore.
CS graduate student Sam Kumar (advisors: David Culler and Raluca Ada Popa) has won the Jay Lepreau Best Paper Award at the 15th USENIX Symposium on Operating Systems Design and Implementation (OSDI) for "MAGE: Nearly Zero-Cost Virtual Memory for Secure Computation." The OSDI, which brings together "professionals from academic and industrial backgrounds in a premier forum for discussing the design, implementation, and implications of systems software," selects three best papers each year after a double-blind review. Co-authored by Prof. David Culler and Associate Prof. Raluca Ada Popa, the paper introduces an execution engine for secure computation that efficiently runs computations that do not fit in memory. It demonstrates that in many cases, one can run secure computations that do not fit in memory at nearly the same speed as if the underlying machines had unbounded physical memory to fit the entire computation. Kumar works in the Buildings, Energy, and Transportation Systems (BETS) research group in the RISE Lab.
"PlushPal: Storytelling with Interactive Plush Toys and Machine Learning," co-authored by CS Masters student Deanna Gelosi (advisor: Dan Garcia), has won the Best Full Paper Award at the Association for Computing Machinery (ACM) Interaction Design for Children (IDC) conference 2021. IDC is "the premier international conference for researchers, educators and practitioners to share the latest research findings, innovative methodologies and new technologies in the areas of inclusive child-centered design, learning and interaction." The paper, which was presented in the "Physical Computing for Learning" conference session, describes PlushPal, "a web-based design tool for children to make plush toys interactive with machine learning (ML). With PlushPal, children attach micro:bit hardware to stuffed animals, design custom gestures for their toy, and build gesture-recognition ML models to trigger their own sounds." It creates "a novel design space for children to express their ideas using gesture, as well as a description of observed debugging practices, building on efforts to support children using ML to enhance creative play." Gelosi's degree will be in the field of Human-Computer Interaction and New Media, and her research interests include creativity support tools, traditional craft and computing technologies, digital fabrication, and equity in STEAM. She is a member of the Berkeley Center for New Media (BCNM), the Berkeley Institute of Design (BID), and the Tinkering Studio--an R&D lab in the San Francisco Exploratorium.
A new AI system, Rapid Motor Adaptation (RMA), enhances the ability of legged robots, without prior experience or calibration, to adapt to, and traverse, unfamiliar terrain in real time. A test robot figured out how to walk on sand, mud, and tall grass, as well as piles of dirt, pebbles, and cement, in fractions of a second. The project is part of an industry-academic collaboration with the Facebook AI Research (FAIR) group and the Berkeley AI Research (BAIR) lab that includes CS Prof. Jitendra Malik as Principal Investigator, his grad student Ashish Kumar as lead author, and alumnus Deepak Pathak (Ph.D. 2019, advisors: Trevor Darrell and Alexei Efros), now an assistant professor at Carnegie Mellon, among others. RMA combines a base policy algorithm that uses reinforcement learning to teach the robot how to control its body, with an adaptation module that teaches the robot how to react based on how its body moves when it interacts with a new environment. “Computer simulations are unlikely to capture everything,” said Kumar. “Our RMA-enabled robot shows strong adaptation performance to previously unseen environments and learns this adaptation entirely by interacting with its surroundings and learning from experience. That is new.” RMA's base policy and adaptation module run asynchronously and at different frequencies so that it can operate reliably on a small onboard computer.
CS Prof. Bin Yu has been awarded an Honorary Doctorate from the University of Lausanne, Switzerland (UNIL). Honoris causa doctorates are often conferred as a way of recognizing individuals who are unaffiliated with an institution but who have contributed to a specific field or to society in general. Yu was cited as "one of the most influential researchers of her time" for her "international reputation," "her character and her openness to others and to the world," and "the breadth and importance of her contributions" which "are far from being confined to the scientific community" and "are part of collective efforts to build a better world." These include her recent work predicting the severity of COVID-19 in the United States. Yu has a shared appointment in the Department of Statistics, and is affiliated with the Berkeley Institute for Data Science (BIDS) and the Berkeley Center for Computational Biology.
CS Prof. Stuart Russell, has been named a 2021 Officer of the Most Excellent Order of the British Empire (OBE). The Officer rank is the second of the order, and is bestowed by the Sovereign of the United Kingdom twice a year to reward valuable "services rendered to the United Kingdom and its people." Russell, who co-authored the world's most popular AI textbook, Artificial Intelligence: A Modern Approach, and founded the Berkeley Center for Human-Compatible Artificial Intelligence (CHAI), was cited for "For services to artificial intelligence research." He is an innovator in probabilistic knowledge representation, reasoning, and learning, including its application to global seismic monitoring for the Comprehensive Nuclear-Test-Ban Treaty. He is also a powerful advocate for the creation of "safe AI" and is active in the movement to ban the manufacture and use of autonomous weapons. His official title is now: Professor Stuart Russell OBE.
Five projects led by EECS faculty have won C3.ai Digital Transformation Institute (DTI) AI for Energy and Climate Security Awards. The awards recognize projects that are using AI techniques and digital transformation to advance energy efficiency and lead the way to a lower-carbon, higher-efficiency economy that will ensure energy and climate security. "C3.ai DTI selects research proposals that inspire cooperative research and advance machine learning and other AI subdisciplines. Projects are peer-reviewed on the basis of scientific merit, prior accomplishments of the principal investigator and co-principal investigators, the use of AI, machine learning, data analytics, and cloud computing in the research project, and the suitability for testing the methods at scale." Each project was awarded $100,000 to $250,000, for an initial period of one year. The winning proposals were:
Offline Reinforcement Learning for Energy-Efficient Power Grids - Sergey Levine, Assistant Professor, Electrical Engineering and Computer Sciences We propose to develop offline RL algorithms to incorporate real-world data in training an RL agent to reduce emissions associated with running an electrical grid.
Sharing Mobile Energy Storage: Platforms and Learning Algorithms - Kameshwar Poolla, Cadence Design Systems Distinguished Professor of Mechanical Engineering This proposal aims to design, validate, and test platforms and learning algorithms for mobile storage applications, which can simultaneously serve the role of generation (supplying energy) and distribution (reticulating energy).
Reinforcement Learning for a Resilient Electric Power System - Alberto Sangiovanni-Vincentelli, Edgar L. and Harold H. Buttner Chair of Electrical Engineering and Computer Science Harnessing the potential of AI techniques to make the power system resilient against such extreme cases is crucial. We propose to develop AI-based methods, and corresponding testing strategies, to achieve this goal.
Affordable Gigaton-Scale Carbon Sequestration: Navigating Autonomous Seaweed Growth Platforms by Leveraging Complex Ocean Currents and Machine Learning - Claire Tomlin, Charles A. Desoer Chair in the College of Engineering A promising approach to carbon sequestration utilizes seaweed, which fixates dissolved CO2 into biomass. Floating platforms that autonomously grow and deposit seaweed could scale this natural process to the open ocean, where the carbon is confined for millennia.
Interpretable Machine Learning Models to Improve Forecasting of Extreme-Weather-Causing Tropical Monster Storms - Da Yang, Faculty Scientist, Lawrence Berkeley National Laboratory, and Bin Yu, Chancellor's Distinguished Professor and Class of 1936 Second Chair Departments of Statistics and Electrical Engineering and Computer Sciences We propose to develop interpretable, machine-learning (ML) models to forecast the Madden-Julian Oscillation (MJO) — the Storm King in Earth’s tropics.
CS Prof. Jennifer Chayes, who is also the Associate Provost for the Division of Computing, Data Science, and Society (CDSS), is the recipient of the 2020 Association for Computing Machinery (ACM) Distinguished Service Award. She was selected for the award, which recognizes outstanding career-long "contributions to the computing community at large," for "her effective leadership, mentorship, and dedication to diversity during her distinguished career of computer science research, teaching, and institution building." Chayes' contributions include leadership at both Microsoft Research (where she founded and led the Theory Group, and Microsoft Research New England, New York City and Montreal) and UC Berkeley (where she is also the Dean of the School of Information); service to many computing and science organizations (including the National Academy of Sciences, the National Research Council, and the ACM A.M. Turing Award Committee); expanding the diversity of the computing field through mentorship of women, underrepresented racial minorities and other disadvantaged groups; and making important research contributions in machine learning.
EECS Prof. S. Shankar Sastry has won the 2021 American Society of Mechanical Engineers (ASME) Rufus Oldenburger Medal for significant contributions and outstanding achievements to the field and profession of automatic control. Sastry, who was dean of Berkeley Engineering for over ten years, was cited “For fundamental contributions to the foundations of nonlinear, adaptive and hybrid control, control of robots and vehicles, and for contributions to control and robotics education.” EECS Prof. Lotfi Zadeh (1921-2017) previously won this award in 1993. The medal will be presented at the ASME Dynamic Systems and Control Division Awards ceremony and dinner, which will take place at the newly instituted Modeling, Estimation and Control Conference (MECC 2021), in Texas in October.
A team at the Berkeley Natural Language Processing Group (NLP) helped augment an AI system named "Dr. Fill" that has won the 2021 American Crossword Puzzle Tournament (ACPT). This is the first time in the contest's history that an AI has trumped its human competitors. The team, which included CS Prof. Dan Klein, graduate students Nicholas Tomlin, Eric Wallace, and Kevin Yang, and undergraduate students Albert Xu and Eshaan Pathak, approached Matthew Ginsberg, who created the Dr. Fill algorithm in 2012, and offered to join forces by contributing their machine learning system called the Berkeley Crossword Solver (BCS). BCS employs a neural network model to combine general language understanding with more "creative" crossword puzzle clues, then applies its knowledge to practice puzzles, improving as it learns. “We had a state-of-the-art natural language understanding and question-answering component but a pretty basic crossword handler, while Matt had the best crossword system around and a bunch of domain expertise, so it was natural to join forces,” said Klein. “As we talked, we realized that our systems were designed in a way that made it very easy to interoperate because they both speak the language of probabilities.” ACPT is the oldest and biggest tournament of its kind, consisting of seven qualifying puzzles and a final playoff puzzle; solvers are ranked using a formula that balances accuracy and speed. Although Dr. Fill made three errors, it completed most puzzles in well under a minute, and ultimately outscored its top human competitor, who made zero errors, by 15 points. The contest was held online this year and attracted more than 1,100 contestants vying for the $3K grand prize.
