Data Science newsletter – October 5, 2017

GROUP CURATION: N/A

Improving swaps data reporting and clearinghouse oversight are CFTC priorities as the agency seeks to rectify certain “unintended consequences” of its post-crisis rulemaking, Chairman J. Christopher Giancarlo said Oct. 3.

Clearinghouses have become “supersized” through mandatory clearing of swaps and need extra attention because of the enormous amount of risk they carry, Giancarlo said at a financial services forum in Washington. Clearing may also be too closely modeled on futures trading and not enough on swaps, the chairman said.

In response to an inquiry about swaps data-reporting issues, Giancarlo said that “of all the post-crisis reforms, visibility into counterparty credit risk is probably the most important. That’s what drove the crisis—a fear that a cascading falling of banks would bring down others because of their connectedness, but a lack of visibility into what that connectedness was,” he told reporters following the George Washington University event. “Nine years after the crisis, we still don’t have that transparency.”

To accompany the new Pixel smartphones announced Wednesday, Google debuted new wireless earbuds, dubbed “Pixel Buds.” These are Google’s first wireless earbuds that are built to be used with Pixel smartphones, but they also give users access to Google Translate so they can have conversations with people who speak a different language.

Genomics researcher Alex Feltus says he never would have considered joining Clemson University’s faculty 10 years ago were it not for one cutting-edge tool there.

The Palmetto Cluster supercomputer.

“They had the computer I needed,” Feltus said. “It’s a basic tool of gene research.”

The supercomputer, housed in a Pendleton research park near Clemson’s Advanced Materials Research Lab, is the eighth most powerful among U.S. academic institutions, according to the supercomputing website Top500.org and could climb higher in those rankings thanks to a $1 million upgrade courtesy of the National Science Foundation (NSF).

Google’s research sibling is bringing in Columbia development professor Jeffrey Sachs, Oxford AI professor Nick Bostrom and climate change campaigner Christiana Figueres to advise the unit.

Yesterday we published ‘Estimated effects of implementing an open access policy for grantees at a private foundation‘ by Carly Strasser and Eesha Khare which looked at the effects of The Gordon and Betty Moore Foundation’s policy to require grantees to publish in open access journals. Research funders are increasingly adjusting their policies to ensure their researchers are sharing their work in publically accessible ways.

The researchers collected data on more than 2,000 publications in over 500 journals that were generated by GBMF grantees since 2001 and examined the journal policies to establish how open access policies might have affected grantee publishing habits. The research found that 99.3% of the articles published by GBMF grantees would have complied with a policy that requires open access within 12 months of publication. Based in part on this study, GBMF has implemented a new open access policy that requires grantees make peer-reviewed publications fully available within 12 months.

Today, we introduced our second generation family of consumer hardware products that are coming to Canada, all made by Google: new Pixel phones, Google Home Mini and Max, an all new Pixelbook, Google Pixel Buds, and an updated Daydream View headset. We see tremendous potential for devices to be helpful, make your life easier, and even get better over time when they’re created at the intersection of hardware, software and advanced artificial intelligence (AI).

While more people are readily introducing these devices into both home and business settings, experts warn of the risks and challenges associated with voice technology. “The addition of voice absolutely increases the risk level for technology users,” said Nathan Wenzler, chief security strategist at AsTech Consulting, a cyberrisk management firm. “When you add more features to a device, you are also adding complexity and more code and, as a result, you are introducing more avenues for people to hack into the device. It’s a major risk component.”

Most devices that employ voice-response technology are internet of things devices and, like many data-collecting devices in this nascent category, manufacturers often do not embed adequate security measures into them. “It can be very easy to break into voice-enabled IoT devices and compromise them, and that opens up a lot of problems,” he said.

Why give away this valuable intellectual property for free? Deepu Talla, Nvidia’s vice president for autonomous machines, says he wants to help AI chips reach more markets than Nvidia can accommodate itself. While his unit works to put the DLA in cars, robots, and drones, he expects others to build chips that put it into diverse markets ranging from security cameras to kitchen gadgets to medical devices. “There are going to be hundreds of billions of internet of things devices in the future,” says Talla. “We cannot address all the markets out there.”

Just over a year ago we presented WaveNet, a new deep neural network for generating raw audio waveforms that is capable of producing better and more realistic-sounding speech than existing techniques. At that time, the model was a research prototype and was too computationally intensive to work in consumer products.

But over the last 12 months we have worked hard to significantly improve both the speed and quality of our model and today we are proud to announce that an updated version of WaveNet is being used to generate the Google Assistant voices for US English and Japanese across all platforms

The two companies teamed up in August to take on Amazon and its Echo.

Michael Cohen, a rising star in the computer science field and a research fellow at the Simons Institute for the Theory of Computing, died last week in Berkeley at the age of 25. Berkeley police said he died of natural causes, but the specific cause of death is still pending.

During Cohen’s time as a graduate student studying computer science at the Massachusetts Institute of Technology, he developed fast algorithms for analyzing high-dimensional data, breaking barriers in the field that had been in place since the 1970s. Cohen also co-authored at least 30 published papers in the past three years.

Unbeknownst to me, at the very moment on Monday morning when I was asking Google CEO Sundar Pichai about the biggest ethical concern for AI today, Google’s algorithms were promoting misinformation about the Las Vegas shooting.

I was asking in the context of the aftermath of the 2016 election and the misinformation that companies like Facebook, Twitter, and Google were found to have spread. Pichai, I found out later, had a rough idea that something was going wrong with one of his algorithms as we were speaking. So his answer, I think it’s fair to say, also serves as a response to the widespread criticisms the company faced in the days after the shooting.

“I view it as a big responsibility to get it right,” he says. “I think we’ll be able to do these things better over time. But I think the answer to your question, the short answer and the only answer, is we feel huge responsibility.” Later, he added, “Today, we overwhelmingly get it right. But I think every single time we stumble. I feel the pain, and I think we should be held accountable.”

With emerging exascale supercomputers, researchers will soon be able to accurately simulate the ground motions of regional earthquakes quickly and in unprecedented detail, as well as predict how these movements will impact energy infrastructure—from the electric grid to local power plants—and scientific research facilities.

Currently, an interdisciplinary team of researchers from the Department of Energy’s (DOE’s) Lawrence Berkeley (Berkeley Lab) and Lawrence Livermore (LLNL) national laboratories, as well as the University of California at Davis are building the first-ever end-to-end simulation code to precisely capture the geology and physics of regional earthquakes, and how the shaking impacts buildings. This work is part of the DOE’s Exascale Computing Project (ECP), which aims to maximize the benefits of exascale—future supercomputers that will be 50 times faster than our nation’s most powerful system today—for U.S. economic competitiveness, national security and scientific discovery.

Intel announced that the new Google Clips* hands-free camera uses the Movidius™ Myriad™ 2 vision processing unit (VPU) for on-device artificial intelligence (AI) processing.

Google Clips captures and curates motion photos of a person’s family, friends and pets, making them accessible in the Clips app.

Studying the human mind is tough. You can ask people how they think, but they often don’t know. You can scan their brains, but the tools are blunt. You can damage their brains and watch what happens, but they don’t take kindly to that. So even a task as supposedly simple as the first step in reading—recognizing letters on a page—keeps scientists guessing.

Now, psychologists are using artificial intelligence (AI) to probe how our minds actually work. Marco Zorzi, a psychologist at the University of Padua in Italy, used artificial neural networks to show how the brain might “hijack” existing connections in the visual cortex to recognize the letters of the alphabet, he and colleagues reported last month in Nature Human Behaviour. Zorzi spoke with Science about the study and about his other work.

When engineers want to make an object weigh less, they literally cut corners. Using a tool called topology optimization, they enlist computers to snip as much material as possible from the inside of objects, reducing the number of spokes on a bicycle wheel, for example. But current methods can only optimize simple objects such as brackets and pipes. Now, a team of researchers says it has created a new method of paring down large-scale objects. The trick is resolution. Three-dimensional images are measured in voxels, a bit like computer images that are measured in pixels. In the past, the resolution of optimized 3D models was limited to 5 million voxels, but the new program—reported today in Nature—can optimize objects up to 1 billion voxels in size. The engineers put the system through its paces by feeding it the wing dimensions from a Boeing 777 airliner. A supercomputer crunched the numbers for 5 days and produced a new design: a wing with a radical internal structure that is kept solid through curved wing spars and diagonal ribs, instead of the gridlike internal ribbing present in standard airplane wings.

By the time a terrorist attack has begun, the security services have already failed. But the challenge they face in detecting potential attacks is substantial, especially since the tactic of terrorism has increasingly been taken up by individual attackers inspired by, but not directly beholden to, formal movements. Spotting a lone wolf among the flock is no easy task, especially when it relies on a bottleneck of human analysis. A new paper in the journal Aggression and Violent Behavior uses a test case of a real lone wolf attack to explore ways we may be able to deal with this in the future. Using online language analysis tools, it hunts within blocks of text for the warning signs we might otherwise miss, with the hope of helping us to more effectively detect the predator.

Giti Zahedzadeh of the Centre for Neuroeconomic Studies at Claremont Graduate University in California wanted to see what insights could be gleaned about violent tendencies by analysing text using the open-source web application Voyant. The application has a number of features such as creating word clouds, word frequency lists, and word association chains to pull out patterns from blocks of text.

All of the diverse operations of Verily, the life sciences technology company that began its life as Google X Life Sciences, can be summed up with a simple sentence: We should be treating our bodies at least as well as we treat our cars.

At the Health 2.0 Fall Conference in Santa Clara this week, Verily Chief Technology Officer Brian Otis spoke with Health 2.0 cofounder and CEO Dr. Indu Subaya on stage.

“We have instrumented cars with a huge number of sensors and communications devices constantly generating data, analyzing that data, and making decisions based on that data,” Otis said. “There are two types of feedback loops that are performed. One is the second-by-second tuning of the engine based on the sensor data. The other is longterm monitoring of maintenance conditions over time. Both of those things could apply to the human body, but we’re just doing a terrible job of it.”

Johns Hopkins Technology Ventures kicked off its inaugural cohort of healthcare startups last month with a focus on connected health and fitness. Called M-1 Ventures, a reference to first-year medical students, the accelerator represents a change of strategy for cultivating early stage companies. Although Johns Hopkins previously collaborated with Dreamit Ventures, it was interested in developing an accelerator on its own. To that end it formed regional alliances with the University of Maryland, Village Capital., and the family office of Baltimore-based Under Armour CEO and Chairman Kevin Plank — Plank Industries.

Other accelerator supporters include UM Ventures —the commercialization arm of the University of Maryland, The Abell Foundation, Brown Advisory, the Maryland Department of Commerce and Village Capital.

Two of the five participants sprang from Johns Hopkins, but the accelerator is open to healthcare entrepreneurs from the Baltimore region. Startups selected for the 16-week program received $25,000 in investment funding and are eligible for additional investments depending on ratings from their peers in the program, according to M-1’s website.

We are pleased to announce that in early 2018, The JAMA Network will launch a new journal—JAMA Network Open. Our editorial goal is to publish the very best clinical research across all disciplines, serving the worldwide community of investigators and clinicians and meeting the evolving needs and requirements of authors and funders. With the launch of JAMA Network Open, we simultaneously assert our editorial commitment to excellence and to the authorship community regardless of requirements of funders. This will be a fully open access journal and follows the launch of JAMA Oncology in 2015 and JAMA Cardiology in 2016, which are hybrid journals offering open access options for research articles.1,2 Frederick P. Rivara, MD, MPH, current editor in chief of JAMA Pediatrics, will be the editor in chief of JAMA Network Open.

Washington, DC The National Academies’ committee on a Decadal Survey of Social and Behavioral Sciences for Applications to National Security will hold three separate public workshops. These workshops will feature invited presentations and discussions to explore the current state of the science and cutting-edge research opportunities. [registration required]

“Sebastian Neubert, a particle physicist at Heidelberg University in Germany, leads a group studying subatomic particles called pentaquarks. The six team members all have access to the software code used to run their multi-step analyses, and the programmers update it daily with new features and bug fixes. With each code change, however, they run the risk of introducing inadvertent errors that foul the underlying algorithms.”

“To prevent that, the team checks and rechecks the analyses, and uses error-checking algorithms, functions they can call whenever a change is proposed, to ensure that their software works as intended. One test, for example, verifies that a noise-cancelling algorithm gives the correct output when it is run on practice data.”

“On the data team here at Stack Overflow, we spend a lot of time and energy thinking about tech ecosystems and how technologies are related to each other. We use these kinds of relationships all over the place, from making the user experience of everyone coming to Stack Overflow better by suggesting relevant content to helping our clients understand how to hire developers. One way to get at this idea of relationships between technologies is tag correlations. Correlation between tags measures how often tags appear together relative to how often they appear separately.”