Data Science newsletter – October 21, 2019

GROUP CURATION: N/A

The AI summit was the fourth InnovationXLab event hosted by a DOE National Laboratory, but it was the first to focus on what Stevens termed an ​“emerging” area of science. ​“The previous summits focused on batteries, the power grid and manufacturing, all of which have had decades of DOE investment and many well-established relationships,” Stevens said. ​“AI, in relative comparison, is so new, and the vast majority of its impact will be felt in the future.”

Unlike fields like batteries or advanced manufacturing, where the national laboratories maintain both deep expertise and tools and capabilities that would be unsustainable by any one company, many AI capabilities are already found in industry, according to Kriplani. ​“The exciting thing about partnerships in AI is that the crossflow of both ideas and capabilities is truly bidirectional,” she said.

Transforming these initial discussions into reality will be helped by the creation of a national AI strategy, which will need to be formulated by a consortium of government, academic and industry partners like those who attended the AI summit, Stevens said. ​“Having an event like this is an essential and effective first step, but it is still a first step,” he said.

… Comprising the first part of tomorrow’s smart environment information-processing pipeline, sensors are the data-gathering apparatus that provide our computers with the information they need to act.

Case Study: The Oura Ring

Not much more than a sleek, black band, the Oura Ring is the most accurate sleep tracker on the market, thanks to its TK sensors.

A Purdue University-affiliated startup, Blue Wave AI Labs, is using predictive analytics to operate nuclear reactors across the United States and to help them operate as safely and efficiently as possible.

Tom Gruenwald, who received his PhD in physics from the Purdue College of Science, and Gina Pattermann, a serial entrepreneur, founded Blue Wave.

Last month, Blue Wave AI Labs received a $6.9 million grant from the Department of Energy to develop predictive models of reactor components that might cause unplanned outages. The Blue Wave research team, based in Purdue Research Park in West Lafayette, will use the funds to analyze billions of nuclear reactor data points.

Below the rumble of passing cars, chirping birds, and rustling leaves, the Earth is constantly humming. This geologic symphony is driven by the ever-sloshing oceans that blanket nearly three-quarters of our planet, but tracing individual refrains from this watery orchestra has long posed a challenge.

Now, researchers have done just that, picking out a previously unknown seismic phenomenon that they have dubbed stormquakes. These events, described this week in Geophysical Research Letters, are pulses of seismic waves birthed from the ferocious energy in massive storms, and they can radiate thousands of miles across continents. (Learn about a different kind of strange seismic wave that rippled around the world.)

“I was surprised by what they saw,” says Göran Ekström, a seismologist at Columbia University who specializes in unusual earthquakes.

Large-scale national research projects hope to reveal how it learns, how it controls behaviour and how it goes wrong.

When ImageNet Roulette first appeared online, people posted pictures of their faces framed by thin neon-green squares, accompanied with words: “swot, nerd, nun, hatmaker”. Scrolling down my newsfeed, I saw what happened when those with darker skin uploaded their photos: “negroid, black person, wrongdoer”. The tool, developed by the US artist Trevor Paglen and researcher Kate Crawford, exposed the labels within the database used for artificial intelligence, and showed how algorithms classify our images using racist tropes.

Imagenet is a vast collection of real pictures used to train machine learning systems (the process commonly known as AI). It houses images for a thousand nouns – apples, oranges, skiing, limousine, Sussex Spaniel, and people; a man in a suit is an “entrepreneur”, a woman in a bikini, smiling, is a “slattern, slut”. Its combinations of words and pictures often feel like like the crowded insides of a feverish brain.

At the curve gallery in the Barbican, London, Paglen, 45, has pinned thousands of pictures on the wall that are taken from ImageNet and classified according to its schema. He starts with an apple, a deliberately benign object, represented according to the numerous apples and apple-like objects that an algorithm has been trained to identify. The categories quickly morph and collide; “Investor” shows a group of suited men; “Picker” groups rural figures with fruit crops and straw baskets; “Crazy” shows humans with a reductive set of horror-movie tropes.

Congestion pricing on our nation’s roadways is forbidden in many cases, or rarely adjusted in others. We’re
still stuck in traffic, with drivers wasting an average of 54 hours every year and costing the economy $166 billion.

What is so maddening is that we know how to fix the problem: Simply price use of the roads based on demand like any other good. There are multiple options for how to implement congestion pricing, some already demonstrated in cities across the world: adding new priced lanes (also known as
managed lanes) to existing freeways;
tolling non-carpool vehicles using high-occupancy vehicle lanes; or
implementing a charge to enter congested central business districts (cordon pricing).

NIH has awarded a $3.7 million grant to Washington University School of Medicine to fund an open-source database that will aim to boost precision medicine and genomic research for cancer.

The Clinical Interpretations of Variants in Cancer (CIViC) database will help doctors match cancer mutations found in patients’ tumors to the appropriate drugs. It is the only entirely open-source online resource for querying cancer mutations.

The system is designed so that anyone can create an account and contribute information to the database, for free. Experts in genomics then edit and curate the data that is incorporated into the system.

Semiconductor manufacturers are trying to seize a share of the growing LiDAR market as autonomous vehicles ramp up. One of these companies is Analog Devices. In a recent phone interview with FierceElectronics, Ron Kapusta, an ADI fellow at Analog Devices, discussed the issues surrounding LiDAR as companies strategize on the best way to move forward with the technology.

Kapusta said that one of the current shortcomings in LiDAR is a shortage of parts optimized for the application. “LiDAR systems are currently almost completely built with components developed for other markets. These technologies are not optimized for the cost and performance needed for LiDAR.”

The Lyle School of Engineering, Guildhall and the Simmons School of Education & Human Development at SMU will use a $1,521,615 grant from the National Science Foundation to research teaching computer science and computational thinking through the popular video game, Minecraft. Research will span the fields of game design, human computer interaction, machine learning, curriculum design and education assessment by integrating STEM+C (computing) based curriculum directly into Minecraft. The project will help advance knowledge in game-based learning by building on techniques and experiences from commercial game design. The game and infrastructure produced through the research will serve as a vital computing resource for middle and high school educators.

One of the most vexing issues I had to deal with as a university president was protecting our students from sexual assault and misconduct. The leaders of America’s leading research universities bear the significant responsibility of providing the safest environment possible for hundreds of thousands of students in both undergraduate programs and graduate and professional schools. This is why I’m so proud that AAU has helped lead the conversation about fighting sexual assault and misconduct on campus.

“Sexual misconduct causes serious harm to those who experience it, is antithetical to the university’s high standards of conduct and is corrosive to Yale’s mission,” Salovey wrote in his Tuesday email. “Such behavior has no place on this campus or anywhere.”

The survey data were analyzed by Westat, an analysis firm, which adjusted the survey results to represent Yale’s population and account for variances in groups’ willingness to report misconduct.

According to Salovey’s email, Yale and 32 other universities participated in the survey, which had a 45.4 percent response rate among Yale’s 13,916 students over the age of 18.

A new report from a University of Illinois panel on faculty misconduct seeks a broader definition of sexual harassment and more transparency. But a university spokesperson couldn’t say when the reforms would be adopted or how much they would cost.

The next evolution in artificial intelligence may be a matter of dispensing with all the probabilistic tricks of deep learning, mastering instead a manifold of shifting energy values, according to deep learning’s pied piper, Yann LeCun.

The bank set aside 35% of its revenue for staff compensation and benefits this year, the lowest that ratio has been in at least a decade, according to an analysis of Goldman’s data.

“As we grow more platform-driven businesses, we expect compensation to decline as a proportion of total operating expenses,” CFO Stephen Scherr says.

London, England November 14, starting at 5 p.m. “You are invited to join the Campaign for Social Science at our annual SAGE Publishing lecture, which this year will be delivered by Mark Easton, Home Editor at the BBC.” [free, registration required]

San Francisco, CA October 16-November 3, 12-8 p.m. daily, 838 Market St. “The Glass Room is a place to explore how technology and data are shaping our perception, experiences, and understanding of the world. The consequences of a ‘move fast and break things’ industry are catching up with us and now we must examine what has been lost and gained along the way. As The Glass Room comes to San Francisco, it holds a mirror up to Big Tech, to reflect its impact on the world and to look for new ways forward.”

Researchers at the San Diego Supercomputer Center (SDSC) at UC San Diego have launched an open-source software called SeedMeLab, which provides a host of features for researchers across all disciplines to manage and disseminate their data products internally and externally from a personalized and branded data cloud with full ownership and control.

Under development and rigorous testing for past five years, SeedMeLab is a result of $1.3 million National Science Foundation (NSF) award to create a set of web-based building blocks that lets scientists seamlessly share and access preliminary results and transient data from research on a variety of platforms, including mobile devices. SeedMe is short for ‘Swiftly Encode, Explore and Disseminate My Experiments’.

“In the 310 series of courses, we are building a neural network framework piece by piece. In Course 311 we built a vizualization for an autoencoder. In Course 312 we built fully connected neural networks. In Course 313 we are adding performance-enhancing features like normalization, dropout, optimizers, custom layers and error functions, and reporting. Course 314 and onward will add hyperparameter optimization, convolutional neural networks, and recurrent neural networks, all with practical examples. It’s unwieldy to call this “the lightweight End-to-End Machine Learning neural network framework”. For convenience, I’ll call it Cottonwood. The code is open source. Try it out for yourself.”

OpenSpiel is a collection of environments and algorithms for research in general reinforcement learning and search/planning in games. OpenSpiel supports n-player (single- and multi- agent) zero-sum, cooperative and general-sum, one-shot and sequential, strictly turn-taking and simultaneous-move, perfect and imperfect information games, as well as traditional multiagent environments such as (partially- and fully- observable) grid worlds and social dilemmas.