Data Science newsletter – December 12, 2018

GROUP CURATION: N/A

We are in desperate need of reevaluating and renaming our modes. 5 or 10 years ago, what kind of data visualization you made — whether it was a dashboard, a notebook, a report or a bespoke communications piece — was very different depending on the language, library or tool you were using. That’s no longer the case.

We need to imagine new approaches which acknowledge that convergence isn’t just happening in the capabilities of tools but also in the expectation of users who no longer are willing to accept that they need to exit out of one mode to optimize for another. That entails shifting our emphasis away from individual charts to the construction, evaluation and delivery of the products where those charts appear.

Every student will need some hands-on experience with the statistical models newly central to our culture. And every student does need to understand that these models are limited by the same contextual provisos that make historians and literary critics so cautious. Tech leaders who argue that machine learning is more objective than other knowledge cannot be trusted. But we should just as fiercely distrust political leaders who use the perspectival complexity of the internet to imply that real knowledge is impossible, everything is fake, and we can only fall back on affinity and prejudice. It is possible to build real knowledge by comparing perspectives from different social contexts. Historians have long known how. As our knowledge about the present is increasingly filtered through statistical models aimed at specific target markets, we may need the same comparative strategies to understand our own lives. More fundamentally, we need to realize that historians’ traditions of caution and relativism are not alien to the sparkling new world of computers.

So what were this year’s highlights? Well, Intel open-sourced a useful tool — HE-Transformer — that allows AI systems to operate on encrypted data. IBM detailed two breakthrough AI training techniques — a digital method that’s up to 4 times faster than the previous state of the art and an analog chip with phase-change memory — that both retain 8-bit precision. And Nvidia described a generative model that can create three-dimensional environments using real-world videos from sources like YouTube.

Without artificial intelligence there wouldn’t be much left of Facebook as we know it today.

That’s according to Yann LeCun, who founded Facebook’s artificial intelligence research lab five years ago.

“If you take the deep learning out of Facebook today, Facebook’s dust,” LeCun, Facebook’s chief AI scientist, recently told CNN Business. “It’s entirely built around it now.”

A few years ago, scientists funded by the Wellcome Trust, one of the world’s wealthiest private philanthropies, published sobering findings about the deadly effects of air pollution. In a long-term study of elderly residents of Hong Kong, China, those exposed to higher levels of smog—especially tiny particles of soot produced by burning fossil fuels—were more likely to die of cancer than people who breathed cleaner air.

The study, published in Cancer Epidemiology, Biomarkers & Prevention in 2016 by researchers from the University of Hong Kong and the University of Birmingham in the United Kingdom, is one of many to highlight the health threats posed by soot. And it is just one product of the extensive investments that Wellcome, with $29.3 billion in assets, has made in environmental science. “We aim to stimulate research excellence and develop global collaborations to drive change,” the London-based philanthropy explains on a web page that highlights its commitment to making “cities healthy and environmentally sustainable.”

The European Commission is preparing to launch a “risk data hub” in the coming months that will help map out loss and damage from natural disasters such as floods, droughts, storms and other extreme weather events that are becoming more frequent with climate change.

The American Association for Justice said Monday it could support key arbitration and preemption language in one section of the AV START (S. 1885) bill, which creates the first federal framework for self-driving technology. Senate Commerce Committee staff that day released a bill draft with language the trial lawyers said would preserve the right to bring a claim under state law and reduced the likelihood the self-driving car industry will be able to force Americans into arbitration.

In a push to advance data quality, the U.S. Food and Drug Administration (FDA) today proposed a new framework to advance the use of real-world evidence across drug and biologic development efforts.

The framework includes strategies for filling in the gaps in other sources of real-world data, an effort that could include the use of mobile technologies, electronic patient-reported outcome tools, wearables and biosensors.

It also outlines real-world evidence-related efforts to explore the potential for using this data to support approval of new indications for approved drugs or to support or satisfy post-approval study requirements.

A Rhode Island Hospital emergency physician and Brown University professor was the lead author for a piece in the New England Journal of Medicine Wednesday about gun violence and the Twitter hashtag #ThisIsOurLane.

Dr. Megan L. Ranney and two co-authors wrote about how doctors and medical professionals have a role in addressing America’s gun violence epidemic. The hashtag #ThisIsOurLane came in response to an NRA tweet that admonished “self-important anti-gun doctors to stay in their lane.”

Boiled down to dollars and cents, the battle between the University of California, the nation’s premier producer of academic research, and Reed Elsevier, the world’s leading publisher of academic journals, can seem almost trivial. UC is paying almost $11 million this year for subscriptions to some 1,500 Elsevier journals. That’s not much when measured against the university’s core budget of $9.3 billion.

But in fact it’s a very big deal — big enough for the university to consider dropping the subscriptions entirely when its current five-year contract with Elsevier expires on Dec. 31. Scores of town hall meetings for UC faculty to discuss the ongoing negotiations between UC and Elsevier have been scheduled across the system as the deadline approaches. What faculty are likely to hear, in the words of Jeff MacKie-Mason, the university librarian at UC Berkeley, is that “we’re pretty far apart at this point.”

That’s because more than money is at stake. The key issue separating the university and the publisher is the concept of “open access.”

Some of Europe’s top machine-learning researchers have founded an organization to strengthen capacity in artificial intelligence (AI) technology on the continent.

The need for such an organization was first described in an open letter written by almost 200 researchers and published in April. The letter warned that investment and expertise in AI technology in Europe is falling behind those in North America and China.

The researchers behind the initiative — called the European Laboratory for Learning and Intelligent Systems (ELLIS) — fear that European universities are losing talent to industry firms, mostly based in the United States and China. Companies in Europe that will require AI to remain competitive are also lagging behind, they say.

Supported by the National Science Foundation, WPI will implement an innovative promotion policy to more equitably support midcareer female faculty and recognize a broader range of faculty work

Rapid research progress in science and technology (S&T) and continuously shifting workforce needs exert pressure on each other and on the educational and training systems that link them. Higher education institutions aim to equip new generations of students with skills and expertise relevant to workforce participation for decades to come, but their offerings sometimes misalign with commercial needs and new techniques forged at the frontiers of research. Here, we analyze and visualize the dynamic skill (mis-)alignment between academic push, industry pull, and educational offerings, paying special attention to the rapidly emerging areas of data science and data engineering (DS/DE). The visualizations and computational models presented here can help key decision makers understand the evolving structure of skills so that they can craft educational programs that serve workforce needs. Our study uses millions of publications, course syllabi, and job advertisements published between 2010 and 2016. We show how courses mediate between research and jobs. We also discover responsiveness in the academic, educational, and industrial system in how skill demands from industry are as likely to drive skill attention in research as the converse. Finally, we reveal the increasing importance of uniquely human skills, such as communication, negotiation, and persuasion. These skills are currently underexamined in research and undersupplied through education for the labor market. In an increasingly data-driven economy, the demand for “soft” social skills, like teamwork and communication, increase with greater demand for “hard” technical skills and tools. [full text]

High-achieving, low-income students who received personalized commitment of financial aid are more than twice as likely to apply, be admitted to and enroll in a top-tier university, according to a new University of Michigan study.

Research by Susan Dynarski, U-M professor of public policy, education and economics, and colleagues found that with this early commitment of aid, high school students were twice as likely to apply to U-M (67 percent compared to 26 percent) and twice as likely to enroll (26 percent compared to 12 percent).

“Students typically find out about financial aid far too late for it to affect their application choices,” Dynarski said. “Our unconditional, early commitment of four years of free tuition and fees assured students and their families that University of Michigan is within their reach.”

Melbourne, Australia February 4, 2019. “The theme of the conference is Collaborations and Partnerships: addressing the big digital challenges together.” [$$$]

Today, we are excited to share TF-Ranking, a scalable TensorFlow-based library for learning-to-rank. As described in our recent paper, TF-Ranking provides a unified framework that includes a suite of state-of-the-art learning-to-rank algorithms, and supports pairwise or listwise loss functions, multi-item scoring, ranking metric optimization, and unbiased learning-to-rank.

PyTables is an efficient method for storing and querying both numerical and textual data. PyTables provides seamless access to the convenient HDF5 library, a popular container for datasets that can grow to terabytes and beyond.