Data Science newsletter – June 28, 2021

A team of Penn State-led researchers recently received a $1.2 million National Science Foundation (NSF) grant to build a search engine and other resources that can make the web safer for users by helping scientists scour billions of online documents to more efficiently collect and classify privacy documentation.

Jointly housed in the College of Engineering and the College of Science, with involvement from the Lowell Institute for Mineral Resources, the school will offer undergraduate, graduate and professional training in areas including data science, business, social sciences, public health and law.

“For the first time in the history of mining anywhere in the world, our new school will rely on a holistic approach to mineral resource management,” said Moe Momayez, interim department head and David and Edith Lowell Chair in the Department of Mining and Geological Engineering. “The Department of Mining and Geological Engineering will play a key role in this effort, just as it has been at the forefront of the field for the past 135 years, and will continue to educate the next generations of mining engineers from every corner of the globe.”

The European Commission may have been hard at work looking for ways to improve the circulation of knowledge across Europe, but to remove barriers between member states, it needs to set up a binding framework for knowledge sharing, says Karen Maex, rector of the University of Amsterdam.

The EU’s plans for a single market for research and for creating a common education area by 2025 are planting the seeds for a unified innovation market. But as things stand, neither of these initiatives requires member states to remove national barriers to knowledge sharing.

“We would need what some call a European Knowledge Act, so that we have a legally binding framework to share our knowledge,” Maex said in a panel discussion on the transformation agenda for universities at the Commission’s Research & Innovations Days event, earlier this week.

The newest co-editor-in-chief of BMC’s Genes & Nutrition, Ahmed El-Sohemy is a Full Professor and Associate Chair at the University of Toronto. Professor El-Sohemy will be speaking at the Personalized Nutrition Innovation Summit on June 28th-29th, and in this new Q&A, Professor El-Sohemy discusses the summit, his company Nutrigenomix, and the journal.

In July 2020, OpenAI launched GPT-3, an artificial intelligence language model that quickly stoked excitement about computers writing poetry, news articles, and programming code. Just as quickly, it was shown to sometimes be foulmouthed and toxic. OpenAI said it was working on fixes, but the company recently discovered GPT-3 was being used to generate child porn.

Now OpenAI researchers say they’ve found a way to curtail GPT-3’s toxic text by feeding the program roughly 100 encyclopedia-like samples of writing by human professionals on topics like history and technology but also abuse, violence, and injustice.

OpenAI’s project shows how the tech industry is scrambling to constrain the dark side of a technology that’s shown enormous potential but also can spread disinformation and perpetuate biases. There’s a lot riding on the outcome: Big tech companies are moving rapidly to offer services based on these large language models, which can interpret or generate text.

The important point of this paper is that studying and influencing collective behaviour has political, ethical and moral dimensions, and needs oversight just as much as e.g. biotech does. I aimed to suggest as much in Critical Mass.

Can a sensor be designed from the ground up, using only computer modeling and simulation?

The concept of computer-aided drug design is now more than 40 years old.(1) Many of the small-molecule drugs prescribed today were discovered using sophisticated computational approaches that predicted and then optimized properties such as binding affinity, absorption by targeted tissues, distribution in the body, metabolism, excretion, and lowered toxicity. In one respect, the challenge faced by those of us designing new sensors is similar to the challenge faced by designers of new therapeutics: the drug (in our case, the analyte) has to engage with its target receptor (the capture probe, the plasmonic metal surface, and the fluorophore, etc.) with high specificity. Additionally, just as a drug has to induce a higher order physiological response in the organism, an analyte has to induce a higher order physical response in the sensor—in the form of a shift in some resonant frequency, a change in impedance, a change in fluorescence, etc. So why are we not as adept at computer-aided sensor design as we are at computer-aided drug design?

I think there are many answers to this question, not the least of which is the larger amount of human effort and financing directed at drug development relative to sensor development.

According to UC Berkeley School of Information professor Deirdre Mulligan, if “Move fast and break things” is the motto of a famous Silicon Valley tech company, then public interest technologists’ might be: “Think critically and consider others.” This summer, Mulligan and fellow I School professor Jenna Burrell hope to show UC Berkeley undergraduates how to do just that with a new program, Fiat Justice Scholars.

Powered by a grant from the New America Public Interest Technology University Network (PIT-UN), the program is a collaboration between the Algorithmic Fairness and Opacity Group (AFOG) at the UC Berkeley School of Information, CalNERDS, and the D-Lab, with the goal of cultivating a pipeline for undergraduate public interest technology scholars at UC Berkeley.

Fiat Justice is unique in that it isn’t cloistered in a STEM program; it’s meant for students of broad backgrounds and skillsets, and the first cohort of twenty reflects this; 82% of participants are first-generation college students, 76% are financial aid recipients, and 55% identify as female. Additionally, roughly 35% are transfer students, 15% are Disabled Students’ Program students, and overall there is strong participation from underrepresented minority groups, a goal of the program. There’s also a diversity of majors like Legal Studies, Molecular & Cell Biology, Public Health, Political Economy, and Urban Studies; roughly 40% of students come from non-STEM majors. The justice-minded program seeks to develop and train diverse students and scholars with the knowledge and skills to create, use, assess, and critique technologies in service of the public interest.

This year was off the charts in most of the western U.S., with a negative anomaly of 20+ days.

Billionaire and Democratic megadonor James Simons has hired a new team of lobbyists as he seeks to influence a federal government now led by President Joe Biden.

Simons supported Biden during the 2020 election.

Simons, according to a new lobbying report, recently hired five powerful lobbyists, including former Republican Senate Majority Leader Trent Lott and former Democratic Sen. John Breaux. Lott, Breaux and Biden served in the Senate together.

Today we are announcing the publication of The Principles of Deep Learning Theory: An Effective Theory Approach to Understanding Neural Networks, a collaboration between Sho Yaida of Facebook AI Research, Dan Roberts of MIT and Salesforce, and Boris Hanin at Princeton. At a fundamental level, the book provides a theoretical framework for understanding DNNs from first principles. For AI practitioners, this understanding could significantly reduce the amount of trial and error needed to train these DNNs. It could, for example, reveal the optimal hyperparameters for any given model without going through the time- and compute-intensive experimentation required today.

The Principles of Deep Learning Theory will be published by Cambridge University Press in early 2022 and the manuscript is now publicly available. “The book presents an appealing approach to machine learning based on expansions familiar in theoretical physics,” said Eva Silverstein, a Professor of Physics at Stanford University. “It will be exciting to see how far these methods go in understanding and improving AI.”

The Texas A&M Institute of Data Science (TAMIDS) has presented grants to 18 Texas A&M faculty members representing the colleges of architecture, engineering, geosciences, liberal arts, medicine, science and University Libraries.

“TAMIDS aims to help faculty strengthen data science in their own departments and foster their leadership of TAMIDS activities. The two TAMIDS programs announced here will sponsor the development of new interdisciplinary courses in data science and support junior faculty in leading workshops, hackathons and K-12 outreach that will engage new groups of students with data science,” said Nick Duffield, director of TAMIDS and holder of the Royce E. Wisenbaker Professorship in the Department of Electrical & Computer Engineering in the College of Engineering.

Across from a dog park in the heart of Washington, D.C., stands a striking, multicolored mural, in which two women reach for each other across a space teeming with variegated particles. The 23-meter-wide mural, inspired by the work of Duke University particle physicist Ayana Arce, who is Black, imagines women building bridges to each other, just as quarks that are unpaired after intense proton-proton collisions find other quarks. Scientist-turned-artist Amanda Phingbodhipakkiya finished the artwork this month; it is the second in a series planned for 10 U.S. cities highlighting the research of female scientists, in a project sponsored by the Heising-Simons Foundation.

Born in Atlanta to Thai and Indonesian immigrants, Phingbodhipakkiya knew she wanted both science and art to be integral to her life. After a life-changing accident derailed a blossoming dance career, she was driven to study neuroscience in college.

But after 4 years as a research assistant in an Alzheimer’s disease lab, she became keenly aware of how poorly scientists–herself included–communicate with the public. So she abandoned her Ph.D. ambitions to get a master’s degree in fine arts (MFA) at the Pratt Institute. Her decision launched a career in science-focused art and design, leading to a TED residency, museum exhibits, and projects that, she says, focus on “badass women in STEM [science, technology, engineering, and math].”

Using a $7.5 million, five-year grant from the U.S. Department of Defense, a multi-university team that includes Johns Hopkins engineers is tackling one of today’s most complex and important technological challenges: How to ensure the safety of autonomous systems, from self-driving cars and aerial delivery drones to robotic surgical assistants.

René Vidal, professor of biomedical engineering and the inaugural director of the Mathematical Institute for Data Science, and Noah J. Cowan, professor of mechanical engineering and director of the LIMBS Laboratory, are partnering with researchers at Northeastern University, University of Michigan, and the University of California, Berkeley, on the Multidisciplinary University Research Initiative, or MURI project, called “Control and Learning Enabled Verifiable Robust AI,” or CLEVR-AI.

A Dutch university says it is formally abandoning the impact factor — a standard measure of scientific success — in all hiring and promotion decisions. By early 2022, every department at Utrecht University in the Netherlands will judge its scholars by other standards, including their commitment to teamwork and their efforts to promote open science, says Paul Boselie, a governance researcher and the project leader for the university’s new Recognition and Rewards scheme. “Impact factors don’t really reflect the quality of an individual researcher or academic,” he says. “We have a strong belief that something has to change, and abandoning the impact factor is one of those changes.”

Online July 14. “We are excited to partner with cnvrg.io on MLCon! It is a FREE virtual ML community conference meant to break down silos, share lessons learned, pro tips, and proven strategies for building real-world AI applications.” [registration required, free]

Data supply chain management, with data products as the end result of the process, can help to address each of these issues. It puts equal emphasis on all phases of data management — from collection to organization to consumption of data products. It’s a means of balancing the benefits of common data with those of unique and tailored data in products, and it’s equally suited to internal and external data. Relatively few companies employ data supply chain management, but those that do tend to report better results.

Esri, the global leader in location intelligence, today announced it is releasing for the first time ever a new high-resolution, 2020 global land cover map as part of the company’s Living Atlas. The map was built using European Space Agency (ESA) Sentinel-2 satellite imagery, and developed using a new machine learning workflow teaming with new Esri Silver Partner Impact Observatory, as well as long-time partner Microsoft.

When the idea of companies having ‘public teams’ or ‘spaces’ was first brought up, my mind went to all the ways it could possibly go wrong. But I was also really curious about whether there was any genuine value that organizations could add for our users. So this sparked our first big research and design effort: a 5-day design sprint, attended by a wide range of teams from across the company; product management, community management, product design, engineering, data, product research, and more.

While Stack Overflow needs paid products in order to operate, we want to do this by adding value to the community, not changing things for the worse. So we settled on the following mission statement for our design sprint: “How might we enable companies to build relationships with the Stack Overflow community in a way that improves Stack Overflow as a resource for developers?”