Grad student Yang You, Prof. James Demmel and Prof. Kurt Keutzer, along with Prof. Cho-Jui Hsieh of UC Davis and Dr. Zhao Zhang of the Texas Advanced Computing Center (TACC), have created, in collaboration with researchers at NVIDIA, a new algorithm which enables them to harness the power of supercomputers to train a deep neural network (DNN) for image recognition at record speed. Deep learning researchers currently use trial-and-error to design new models, requiring them to run training processes tens or even hundreds of times for each model.  The team's effort efficiently used 1024 Skylake processors on the Stampede2 supercomputer at TACC to complete a 100-epoch ImageNet training with AlexNet in 11 minutes - the fastest time recorded to date.  Also, using 1600 Skylake processors, they bested Facebook's prior results by finishing a 90-epoch ImageNet training with ResNet-50 in 32 minutes and, for batch sizes above 20,000, their accuracy was much higher than Facebook's.   The group's breakthrough involved the development of the Layer-Wise Adaptive Rate Scaling (LARS) algorithm that is capable of distributing data efficiently to many processors to compute simultaneously using a larger-than-ever batch size (up to 32,000 items). The LARS algorithm was jointly developed with Nvidia. The findings show an alternative to the trend of using specialized hardware - either GPUs, Tensor Flow chips, FPGAs or other emerging architectures—for deep learning. The team wrote the code based on Caffe and utilized Intel-Caffe, which supports multi-node training. The results are published in Arxiv.

Prof. Stuart Russell teamed up with Autonomousweapons.org to create a provocative video called Slaughterbots, which shows what can go wrong when autonomous weapons are designed to target humans.  "This short film is more than just speculation" Russell says. "It shows the results of integrating and miniaturizing technology that we already have."

The School of Information (I School) is now accepting applications for its new, online Master of Information and Cybersecurity (MICS) program, also known as cybersecurity@berkeley.  MICS is an accredited program designed to train students with professional experience for a career in cybersecurity. Delivered online, MICS provides the technical skills and contextual knowledge students need to assume leadership positions in private sector technology companies as well as government and military organizations.  The program was developed in partnership with the College of Engineering and in collaboration with the Center for Long-Term Cybersecurity (CLTC).  A number of EECS faculty are involved with the CLTC, which seeks to create effective dialogue among industry, academia, policy makers, and practitioners around a future-oriented conceptualization of cybersecurity — what it could imply and mean for human beings, machines, and the societies that will depend on both.  The first application deadline is January 30, 2018, and the first cohort of students will begin classes in May 2018.

UC Berkeley is ranked #3 overall in the U.S. and global computer science rankings (CSRankings) for 2017.  CSRankings is designed as a more meaningful and transparent alternative to the US News and World Report computer science ranking system--which is entirely reputation-based and relies on surveys sent to department heads and directors of graduate studies.  The CSRankings system is entirely metrics-based: it weighs departments by their presence at the most prestigious publication venues.    Berkeley ranked in the top 10 in all 4 fields:  Theory (1), Artificial Intelligence (3), Systems (6) and Interdisciplinary Areas (6).  And of the 26 areas ranked, Berkeley made the top 10 in 11 of them: computer vision(2), robotics(2), machine learning and data mining(3), computer security(3), cryptography(3), design automation(3), operating systems(4), natural language processing(5), software engineering(6), algorithms & complexity(7), computer networks(8).

Start-up Embodied Intelligence,  founded by Prof. Pieter Abbeel and his grad students Peter Chen, Rocky Duan, and Tianhao Zhang, raised $7M in a seed round yesterday led by venture capital firm Amplify Partners.  VC firms Lux Capital, SV Angels, FreeS, 11.2 Capital, and A. Capital also supplied capital.  Embodied Intelligence is building AI software to enable robots to learn tasks performed by the user via a virtual reality headset.  It claims existing robots will be compatible with the "robot brain," which would supplant coding scripts tailored to each task.  Embodied will use the seed capital to write its first robotics applications.

CS students enrolled in CS 194, an experimental “cyberwar” course led by Prof. Doug Tygar,  have joined forces with the white hat hackers at HackerOne, a vulnerability coordination and bug bounty platform.  This is the first time that HackerOne--which connects hackers with tech companies, private businesses and federal agencies to hunt for vulnerabilities--has partnered with a university.  Students are gaining real-world cyberwar experience. “Unless students can learn to ‘think like a hacker,’ they will not be able to effectively defend systems” says Tygar.

The start-up Embodied Intelligence and its founders, Prof. Pieter Abbeel and grad students Peter Chen, Rocky Duan, and Tianhao Zhang, are the focus of two news articles: one from the New York Times titled "A.I. Researchers Leave Elon Musk Lab to Begin Robotics Start-Up," and one from Berkeley News titled "Berkeley startup to train robots like puppets."  The start-up is backed by $7 million in funding from Amplify Partners and other investors and will specialize in complex algorithms that allow machines to learn new tasks on their own through extreme trial and error.  The researchers are augmenting the algorithms with a wide range of techniques, like using virtual reality tools to show a robot how to perform a task--translating the movements into digital data.  “With our advances in machine learning, we can write a piece of software once — machine learning code that enables the robot to learn — and then when the robot needs to be equipped with a new skill, we simply provide new data.” Abbeel explains.

Six EECS faculty are profiled in a Berkeley Engineering article titled "Case studies in forward thinking."  Pieter Abbeel's vision of the future is one of goal-oriented AI, where machines are learning responsibly.  Claire Tomlin envisions a future of airspace management, where intelligent robots can quickly and safely react to dynamic situations, and maybe even deliver packages on time.   Alexandre Bayen sees transit data trends, where data is used to reveal the previously unobservable.  Ken Goldberg envisions a future of dexterous robots, where machines work together with humans to refine their respective skills sand expertise.  Ren Ng's vision of the future is one where better optics--integrating both hardware and software--improve everything from computer vision and medical diagnostics to family photos and immersive entertainment.   Ana Claudia Arias sees a future which includes electronics and diagnostic equipment that fits the body of a patient and is capable of quickly producing high resolution images, all the while providing a more comfortable experience, particularly for children.

Teaching Prof. Dan Garcia is quoted in a WBRC Fox 6 News article  which discusses how a group of Alabama teachers are working to expand computer science education opportunities for students in rural Alabama and inner-city Birmingham.  Garcia, who was part of the 2017 Alabama Teachers Computer Science Summit at The University of Alabama, praised efforts of teachers and advocates in Alabama and across the country, for their work to expand CS education. "Graduation day is the happiest day of my life, when I see all of the people who took my course four years ago, and got hooked on it," Garcia said. "They can do anything. Every single industry is being affected by data."

Prof. Pieter Abbeel is featured in a New York Times article titled "Building A.I. That Can Build A.I.," about how Google and others, in competition for a small pool of qualified researchers, are looking for automated ways to deal with a shortage of artificial intelligence experts.   The key might be to build algorithms that analyze the development of other algorithms, learning which methods are successful and which are not--leading to  more effective machine learning.  This could help companies build systems with artificial intelligence even if they don’t have extensive expertise in that area.  Abbeel and his team demonstrate techniques that could allow robots to learn new tasks based on what they have learned in the past. “Computers are going to invent the algorithms for us, essentially,” said  Abbeel. “Algorithms invented by computers can solve many, many problems very quickly — at least that is the hope.”