Data Science newsletter – September 9, 2019

GROUP CURATION: N/A

Tony Jebara joined the streaming music service Spotify last month as vice president of engineering for personalization to help improve Spotify’s recommendations to users and other features fueled by A.I. and machine learning.

Jebara, who is also a computer science professor at Columbia University and the co-author of numerous papers on A.I., said on his LinkedIn page that he’d be leading Spotify’s company-wide machine learning strategy.

Mastercard believes women, who by 2028 will control nearly 75% of consumer discretionary spending worldwide, can play a pivotal role in encryption, fraud detection, biometrics and data analysis.

Through its Girls4Tech program, Mastercard sends employees out to schools worldwide to inspire girls to seek a future in these high-tech fields. Their goal: to reach 1 million girls by 2025.

Egon Willighagen is a chemist at Maastricht University with a passion for open science and standards. In his wallet, on what at first glance looks like a credit card, are the FAIR principles. Research output should, it says, be findable, accessible, interoperable, and reusable. What that means in practice is that articles and deposited data must be interlinked, and data points stored within must be machine readable and usable.

Ultimately, Willighagen says, the FAIR principles are about making sure that the work that chemists and other scientists are doing can be found, extracted, and then applied elsewhere. But despite this seemingly sensible goal, he adds, many scientists have not kept up with ensuring that their raw data are preserved and accessible. “It’s really sad,” he says.

Washington University is spearheading a new effort to diversify the field of data science.

Beginning in 2020, the university will train faculty and grad students from across the country in how to use data science tools and methods. The three-year program will focus specifically on recruiting underrepresented minorities, including Latino, indigeneous and black scholars.

The program aims to equip researchers with the skills they need to be successful in this rapidly changing field, said Odis Johnson Jr., professor of sociology and education at Wash U.

[Tom] Erickson was announced Thursday as the first director of the new School of Computer, Data and Information Sciences, which will include the existing Department of Computer Sciences, Department of Statistics and the Information School. It will also have ties to the American Family Insurance Data Science Institute, a research center created in the spring.

What makes Erickson a non-traditional but “right for the moment” pick to direct the school is not his academic background but his experience in enterprise software and the evolving wave of innovation that stands to transform the industry.

In the process, more UW-Madison students will get at least a taste of computer science, campus researchers from a mix of disciplines will find potential partners, and private industry in Wisconsin will be able to learn how trends such as artificial intelligence, machine learning and data collection and analysis can make them more competitive.

Felicity Huffman told a judge on Friday that she arranged to cheat on her daughter’s SAT exam out of a misguided belief that she was giving her daughter a “fair shot” at college acceptance.

Huffman pleaded guilty in April to paying $15,000 to consultant Rick Singer in order to boost her daughter’s SAT result, as part of a much broader federal probe into corruption in college admissions. She is due to be sentenced next Friday in Boston.

Federal prosecutors had initially sought a term of at least four months in prison.

Legal research companies are selling surveillance data and services to U.S. Immigration and Customs Enforcement (“ICE”) and other law enforcement agencies. This Article discusses ethical issues that arise when lawyers buy and use legal research services sold by the same vendors responsible for building ICE’s surveillance systems. As the legal profession collectively pays millions of dollars for computer assisted legal research services, lawyers should consider whether doing so in the era of big data policing compromises their confidentiality requirements and their obligation to supervise third party vendors. With new companies developing legal research services, lawyers have more legal research options than ever. Lawyers can choose to purchase legal research services from socially responsible vendors.

Despite active learning being recognized as a superior method of instruction in the classroom, a major recent survey found that most college STEM instructors still choose traditional teaching methods. This article addresses the long-standing question of why students and faculty remain resistant to active learning. Comparing passive lectures with active learning using a randomized experimental approach and identical course materials, we find that students in the active classroom learn more, but they feel like they learn less. We show that this negative correlation is caused in part by the increased cognitive effort required during active learning. Faculty who adopt active learning are encouraged to intervene and address this misperception, and we describe a successful example of such an intervention.

Matter still matters. And on the 150th anniversary of the periodic table’s formulation by the Russian chemist Dmitri Mendeleev, it’s more important than it’s ever been.

True, technology has made the economy more virtual, but it’s also vastly increased the capability and sophistication of material objects.

Rather than having to painstakingly create the webs by hand as they’d done in the past, technical director Hosuk Chang created a swarm of AI spiders that could weave the webs just like a real spider would.

A first-of-its-kind cyberattack on the U.S. grid created blind spots at a grid control center and several small power generation sites in the western United States, according to a document posted yesterday from the North American Electric Reliability Corp.

The unprecedented cyber disruption this spring did not cause any blackouts, and none of the signal outages at the “low-impact” control center lasted for longer than five minutes, NERC said in the “Lesson Learned” document posted to the grid regulator’s website.

But the March 5 event was significant enough to spur the victim utility to report it to the Department of Energy, marking the first disruptive “cyber event” on record for the U.S. power grid (Energywire, April 30).

You gave them your data in exchange for a driver’s license. DMVs are making tens of millions of dollars selling it, documents obtained by Motherboard show.

A Bellevue-based insurance startup that uses artificial intelligence to sell personalized insurance options has been sold for $2.35 billion to industry giant Prudential Financial.

Assurance IQ, an online insurance-tech company that started in 2016, meets its customers through an online portal instead of through a financial adviser. It sells life insurance, auto insurance, health insurance and Medicare options online from over 20 providers. The company says it has sold at least one insurance product to 300,000 customers and the site has had more than 18 million visitors.”

The gerrymandering trial proceeded on a recent day in the state capital of Raleigh, in a law school room borrowed by the Wake county superior court for the purpose, two blocks from the capitol grounds, where a trio of monuments to the Confederacy, including a 75ft tall obelisk dedicated “to our Confederate dead”, have been the subject of protest recently, but seem unlikely to come down anytime soon.

To say that the trial proceedings were electric would be accurate – for gerrymandering buffs. For three hours, mathematics professor Wesley Pegden testified about how he had created a program to randomly generate trillions of maps of North Carolina state legislative districts that were slightly different from the enacted maps.

With a projected animation on the courtroom walls of a computer furiously redrawing the enacted map, as if it were trying to defeat itself at tic-tac-toe, Pegden demonstrated that only one out of every 100,000 or so hypothetical maps was better for Republicans than the enacted map.

Media researcher and publisher Jeff Pooley responds to the European open access initiative, Plan S, outlining the history of the current system of author-paid article processing charges (APCs) and the ways this system perpetuates inequality across the publishing landscape. He proposes an alternative system wherein university libraries shoulder the publishing costs, and describes an economic framework that could make such a solution sustainable for authors, libraries, publishers, and scholarly societies.

New York, NY October 28-29, at Cornell Tech and Weill Cornell Medicine. “This two-day event sponsored by Cornell’s Academic Integration Initiative aims to bridge the strengths in Ithaca with the clinical and technological environment in New York City.” [registration required]

A constant question that comes up for academics is: Should I accept that review request? When I look back on how I managed my time as a junior faculty member, this was one area where I wasted time by saying yes way too often. Pre-tenure, I nervously said yes to all requests. I was afraid to say no and didn’t understand when it was prudent to say no. Over time, I’ve developed some rules of thumb for when to say yes that I’d like to share.

First (and this is obvious but it needs to be said), I only accept review requests where the content is in my area of expertise. Sometimes part of the content is in my area and part is not, and I will note that in my review. (“I’m not an expert on machine learning, and will comment mainly on the HCI parts of this paper.”)

Second, if I have submitted to a venue (conference, journal), then I owe them reviews back. I have generated a need for reviews, and I need to give back.

Pruning and matrix decomposition are popular ways to reduce model size, but we sometimes find that certain revolutionary neural architecture design choices have a significant impact as well. This is certainly the case with the depth-wise separable convolution popularized in the MobileNet networks. The authors of this paper also found that using depth-wise separable convolutions was the best way to get the most accurate, fastest, and smallest keyword-spotting model when compared to several other techniques.

But can we go further?