research

Google TPUs are built for inference

CS Prof. Emeritus David Patterson co-authored and presented a report on Tensor Processing Units (TPUs) at a regional seminar of the National Academy of Engineering, held at the Computer History Museum in Menlo Park on April 5, 2017.   TPUs, which have been deployed in Google datacenters since 2015, are printed-circuit cards which are inserted into existing servers and act as co-processors tailored for neural-network calculations.  Prof. Patterson says that TPUs are "an order of magnitude faster than contemporary CPUs and GPUs" with an even larger relative performance per watt.  According to an article for the IEEE Spectrum, TPUs are "built for doing inference," having hardware that operates on 8-bit integers rather than the higher-precision floating-point numbers used in CPUs and GPUs.

Meet Ray, the Real-Time Machine-Learning Replacement for Spark

CS Prof. Michael Jordan, graduate students Philipp Moritz and Robert Nishihara, and research in the RISELab are featured in a Datanami article titled "Meet Ray, the Real-Time Machine-Learning Replacement for Spark."  Ray is one of the first technologies to emerge from RISELab, the successor to AMPLab and its host of influential distributed technologies including Spark, Mesos, and Tachyon. Ray is a new distributed framework designed to enable Python-based machine learning and deep learning workloads to execute in real-time with MPI-like power and granularity. This framework is ostensibly a replacement for Spark, which is seen as too slow for some real-world AI applications.

Paper authored by EECS alumni receives 2017 NSDI Test-of-Time Award.

The paper “X-Trace: A Pervasive Network Tracing Framework”, authored by EECS alumi Rodrigo Fonseca (Ph.D. ’08) and George Porter (Ph.D. ’08) and Professors Randy Katz, Scott Shenker, and Ion Stoica, has received the 2017 Networked Systems Design and Implementation (NSDI) Test-of-Time Award. X-Trace was not the first tracing framework, but it was influential given that it was effectively the first framework for end-to-end tracing to focus on generality and pervasiveness. The researchers implemented X-Trace in protocols and software systems, and in their prize-winning paper, they set out to explain three different use scenarios: domain name system (DNS) resolution; a three-tiered photo-hosting website; and a service accessed through an overlay network.

Charles Bordenave awarded the Prix Marc Yor by the SMAI, France

Charles Bordenave, an EECS/Statistics postdoc (Sept. 06-07) has been awarded thePrix Marc Yor by the Société de Mathématiques Appliquées et Industrielles (SMAI) of France. Charles Bordenave was an EECS/Statistics postdoc co-supervised by Prof. Venkatachalam Anantharam (EECS) and Prof. David Aldous (Statistics) and is currently with the French National Center for Scientific Research, the largest governmental research organization in France. This award is given to people under the age of 40 who have practiced in France for at least 5 years. Bordenave is recognized for his works of great scope, creative and stimulating, whose contributions to the theories of random graphs and large random matrices are brilliant and profoundly original.

Tsu-Jae King Liu talks chip efficiency on Moore's Law Panel

EE Prof. Tsu-Jae King Liu, who participated in a semiconductor "Moore's Law" panel discussion at the South by Southwest festival in Texas, is featured in an Electronic Design article  titled "Forget Scaling. Moore's Law Panel Talks Power Consumption."  Tsu-Jae, who helped pioneer the Finfet transistor in 1998, describes some of the ways that transistors and integrated circuits will be evolving and how they might be used in future innovations.

Matthias Vallentin and Vern Paxson take a “VAST” Step Forward in Cyber Security

Postdoctoral researcher Matthias Vallentin is developing VAST,  a  forensic analysis tool  designed to help prioritize the investigation of computer security breaches.  It complements Bro, a security tool  devised by Prof. Vern Paxson when he was a graduate student 22 years ago and which is now used worldwide, to instantly collect huge volumes of log data that a hack might compromise.  “Maybe the external machine also appeared in a phishing email, which contained a PDF attachment. Not only that, but the PDF also includes a malicious payload, which upon opening, sends sensitive information from the employee’s computer to a cyber criminal.  VAST supports this iterative process to reconstruct the complete picture and presents it on a platter” explains Vallentin.  The function, development, and industrial potential of these tools are discussed in a Berkeley Research article.

Baiyu Chen awarded top prize at Infrastructure Vision 2050 Challenge

CS graduate student Baiyu Chen (advisor: Alexei Efros) and Anthony Barrs were awarded top prize and $50,000 for their design at the Infrastructure Vision 2050 Challenge.  Their idea, profiled in an article for Fortune, was to construct a "Hyperlane," or a single platform the size of four interstate lanes that would run parallel to pre-existing highways in order for self-driving cars to travel at high speeds with no chance of getting into a jam.

Tomás Vega rises to a disability challenge

Tomás Vega, a senior computer science and cognitive science major, was part of a team that participated in the first collegiate Tikkun Olam Makers (TOM) event and which was profiled in a Jewish News article titled "Cal students pull a marathon to engineer disability solutions."  His team, led by Berkeley alumnus Pierluigi Mantovani, designed a special pair of gloves to help a Berkeley filmmaker with very limited use of his hands navigate a computer screen.  The filmmaker, who had to use a joystick with his lower lip to navigate, can now perform the same task by just slightly moving his wrists.  The gloves use electromyography, which detects signals from his muscles.   “When someone tells you, ‘Thank you for changing my life, for improving the quality of my life,’ there’s nothing like that,” Vega said.

Dan Garcia's research group GameCrafters featured in PBS Newshour article

Prof. Dan Garcia was interviewed in a PBS Newshour article titled “This Pi Day, use math to beat your friends at classic toy games”. To celebrate Pi Day, mathematicians and physicists explored three classic toys, hula hoops, yo-yo’s and Connect Four on ways to be the most successful with each toy. To count how many possible ways a real game of Connect Four could end, Prof. Garcia’s  research group called GamesCrafters created a program to simulate all the moves. GamesCrafters “strongly solved” the game and found 4,531,985,219,092, or 4.53 trillion combinations to Connect Four are possible.

Jose Carmena wants to use brain implants to tune the mind

EE Prof.  and co-director of the Center for Neural Engineering and Prostheses Jose Carmena participated in a discussion with the Kavli Foundation about technologies that might be used to treat mental illness or enhance memory. Prof. Carmena, who has co-developed brain-machine interfaces and implantable microsensors, says "we have learned a lot about how to target different areas in the brain.  The aim of these devices is to help people who are paralyzed by injury or disease move again by creating an artificial pathway between the areas of the brain that control motion and the muscles. Essentially, they are devices that use electrodes implanted in the brain to translate thoughts into the action of prosthetics."