‘Off label’ use of imaging databases could lead to bias in AI algorithms, study finds

A paper with lead author EECS postdoc Efrat Shimron and co-authors EECS graduate student Ke Wang, UT Austin professor Jonathan Tamir (EECS PhD ’18), and EECS Prof. Michael Lustig shows that algorithms trained using "off-label" or misapplied massive, open-source datasets are subject to integrity-compromising biases.  The study, which was published in the Proceedings of the National Academy of Sciences (PNAS), highlight some of the problems that can arise when data published for one task are used to train algorithms for a different one.  For example, medical imaging studies which use preprocessed images may result in skewed findings that cannot be replicated by others working with the raw data.  The researchers coined the term “implicit data crimes” to describe research results that are biased because algorithms are developed using faulty methodology. “It’s an easy mistake to make because data processing pipelines are applied by the data curators before the data is stored online, and these pipelines are not always described. So, it’s not always clear which images are processed, and which are raw,” said Shimron. “That leads to a problematic mix-and-match approach when developing AI algorithms.”

Tiny switches give solid-state LiDAR record resolution

A new type of high-resolution LiDAR chip developed by EECS Prof. Ming Wu could lead to a new generation of powerful, low-cost 3D sensors for autonomous cars, drones, robots, and smartphones. The paper, which appeared in the journal Nature, was co-authored by his former graduate students Xiaosheng Zhang (Ph.D. '21) and Johannes Henriksson (Ph.D. '21), current graduate student Jianheng Luo, and postdoc Kyungmok Kwon, in the Berkeley Sensor and Actuator Center (BSAC).  Their new, smaller, more efficient, and less expensive LiDAR design is based on a focal plane switch array (FPSA) with a resolution of 16,384 pixels per 1-centimeter square chip, which dwarfs the 512 pixels or less currently found on FPSA.  The design is scalable to megapixel sizes using the same complementary metal-oxide-semiconductor (CMOS) technology used to produce computer processors.   Additionally, large, slow and inefficient thermo-optic switches are replaced by microelectromechanical system (MEMS) switches, which are traditionally used to route light in communications networks.  If the resolution and range of the new system can be improved, conventional CMOS production technology can be used to produce the new, inexpensive chip-sized LiDAR.

Alistair Sinclair and Shafi Goldwasser win inaugural STOC Test of Time awards

CS Profs. Alistair Sinclair and Shafi Goldwasser have won inaugural Test of Time awards at the 2021 Symposium on Theory of Computing (STOC), sponsored by the ACM Special Interest Group on Algorithms and Computation Theory (SIGACT).  Sinclair won the 20 Year award for his paper, “A polynomial-time approximation algorithm for the permanent of a matrix with non-negative entries," which solved a problem that had been open for decades. Goldwasser won the 30 Year award for "Completeness theorems for non-cryptographic fault-tolerant distributed computation," which showed how to compute a distributed function even if up to one-third of the participants may be failing, misbehaving, or malicious.  The awards were presented at the 2021 STOC conference in June.

Shafi Goldwasser wins 2021 FOCS Test of Time Award

CS alumna and Prof. Shafi Goldwasser (Ph.D. '84, advisor: Manuel Blum) has won the 2021 Foundations of Computer Science (FOCS) Test of Time Award.  This award "recognizes papers published in past Annual IEEE Symposia on Foundations of Computer Science (FOCS) for their substantial, lasting, broad, and currently relevant impact. Papers may be awarded for their impact on Theory of Computing, or on Computer Science in general, or on other disciplines of knowledge, or on practice."  Goldwasser is among five co-authors who won the award in the 30 year category for their groundbreaking complexity theory paper "Approximating Clique is Almost NP-Complete," which used the classification of approximation problems to show that some problems in NP remain hard even when only an approximate solution is needed. 

He Yin and Murat Arcak win 2019-20 Brockett-Willems Outstanding Paper Award

EECS Prof. Murat Arcak and his graduate student He Yin have won the second Systems & Control Letters (SCL) Brockett-Willems Outstanding Paper Award. Their paper, "Reachability analysis using dissipation inequalities for uncertain nonlinear systems," published in SCL Volume 142, on August 2020, was deemed the best of 295 papers submitted to the journal in the two-year period between January 2019 through December 2020.  Co-authors include former ME Prof. Andrew Packard, who died in 2019, and Packard's former graduate student, Peter Seiler.  SCL hopes to present the award at the 25th International Symposium on Mathematical Theory of Networks and Systems (MTNS) which will be held in Bayereuth, Germany, in September 2022.

Tsu-Jae King Liu

Tsu-Jae King Liu says the U.S. must revitalize semiconductor education and training

EECS Prof. and dean of Engineering Tsu-Jae King Liu has written an opinion piece for the Mercury News in which she explains why "the country urgently needs to reinvest in semiconductor design and manufacturing, including the development of a highly trained workforce."  She argues that America's lack of a skilled semiconductor manufacturing workforce, in the face of a global semiconductor chip shortage, is a matter of national security because it leaves the country vulnerable to geopolitical instability. "Systems that we rely upon for communications, commerce, defense and more are in jeopardy because the United States has lost its leadership in semiconductor manufacturing over the past three decades."  She appeals to Congress to address the issue and says "we need to double the number of students trained in microelectronics graduating today from all U.S. colleges and universities."  This will require "universities across the nation to collaborate with each other and to partner with industry" to create a geographically-distributed American Semiconductor Academy "with participating schools sharing curricula, facilitating access to industry-leading software tools and coordinating hands-on training for students."

Google Doodle honors Lotfi Zadeh, father of fuzzy logic

EECS Prof. Emeritus Lotfi Zadeh (1921 - 2017) is being honored with a Google Doodle feature today.  In 1964, Zadeh conceived a new mathematical concept called fuzzy logic which offered an alternative to rigid yes-no logic in an effort to mimic how people see the world.  He proposed using imprecise data to solve problems that might have ambiguous or multiple solutions by creating sets where elements have a degree of membership. Considered controversial at the time, fuzzy logic has been hugely influential in both academia and industry, contributing to, among other things, "medicine, economic modelling and consumer products such as anti-lock braking, dishwashers and elevators."   Zadeh's seminal paper, "Fuzzy Sets -- Information and Control," was submitted for publication 57 years ago today.

Michael Jordan calls for a more practical and advantageous approach to AI

CS Prof. Michael Jordan has co-written an article in Wired titled "The Turing Test Is Bad for Business" in which he argues that now that "computers are able to learn from data and...interact, infer, and intervene in real-world problems, side by side with humans," humans should not try to compete with them but "focus on how computers can use data and machine learning to create new kinds of markets, new services, and new ways of connecting humans to each other in economically rewarding ways."  Jordan wrote the article because many AI investors are focusing on technologies with the goal of exceeding human performance on specific tasks, such as natural language translation or game-playing. “From an economic point of view, the goal of exceeding human performance raises the specter of massive unemployment,” he said. “An alternative goal for AI is to discover and support new kinds of interactions among humans that increase job possibilities.”

Rose Abramson wins EPE 2021 Young Author Best Paper Award

EECS graduate student Rose A. Abramson (advisor:  Robert Pilawa-Podgurski) has won the European Power Electronics and Drives Association (EPE) 2021 Young Author Best Paper Award.   Her paper, “A High Performance 48-to-8 V Multi-Resonant Switched-Capacitor Converter for Data Center Applications,” co-authored by EECS alumnus Zichao Ye (Ph.D. '20) and Prof. Robert Pilawa-Podgurski, was presented during the EPE 2020 ECCE Europe conference.  Abramson, whose research focuses on power electronics and energy, received her B.S. in 2015 and her M.Eng. in 2016, both from MIT, and worked as a project electronics engineer at both Nucleus Scientific and Lutron Electronics before coming to Berkeley.   EPE Awards honor outstanding achievements in power electronics and more generally in the field of EPE activities.

Xiaoye Li and Richard Vuduc win 2022 SIAG/SC Best Paper Prize

CS alumni Xiaoye Sherry Li (Ph.D. '96, advisor: James Demmel) and Richard Vuduc (Ph.D. '03, advisor: James Demmel) have, along with Piyush Sao of Georgia Tech, won the 2022 Society for Industrial and Applied Mathematics (SIAM) Activity Group on Supercomputing (AG/SC) Best Paper Prize.  This prize recognizes "the author or authors of the most outstanding paper in the field of parallel scientific and engineering computing published in English in a peer-reviewed journal." Their paper, "A communication-avoiding 3D algorithm for sparse LU factorization on heterogeneous systems,” was published in 2018 in the IEEE International Parallel and Distributed Processing Symposium (IPDPS).  Li is now a Senior Scientist at Lawrence Berkeley National Laboratory (LBNL) where she works on diverse problems in high performance scientific computations, including parallel computing, sparse matrix computations, high precision arithmetic, and combinatorial scientific computing.  Vuduc, now an Associate Professor in the School of Computational Science and Engineering at Georgia Tech, is interested in high-performance computing, with an emphasis on algorithms, performance analysis, and performance engineering.