Data Science newsletter – October 14, 2021

It’s about time we got spooked about the environmental impact of Halloween.

American consumers are expected to spend $3.32 billion on Halloween costumes this year. Most of these will be cheap, disposable costumes sold by retailers like Spirit Halloween and Walmart. Around 83% of materials used to make these costumes is derived from plastic. And many will be thrown out on November 1. In the U.K., one study found that 7 million costumes are thrown out each year—potentially the equivalent of 83 million plastic bottles. In the U.S. which has five times the population and where Halloween is a more popular holiday, that figure will be significantly higher.

The good news is that you can still have a fun-filled Halloween without adding to this waste.

Soon after ratifying our Constitution, Americans adopted a Bill of Rights to guard against the powerful government we had just created—enumerating guarantees such as freedom of expression and assembly, rights to due process and fair trials, and protection against unreasonable search and seizure. Throughout our history we have had to reinterpret, reaffirm, and periodically expand these rights. In the 21st century, we need a “bill of rights” to guard against the powerful technologies we have created.

Our country should clarify the rights and freedoms we expect data-driven technologies to respect. What exactly those are will require discussion, but here are some possibilities: your right to know when and how AI is influencing a decision that affects your civil rights and civil liberties; your freedom from being subjected to AI that hasn’t been carefully audited to ensure that it’s accurate, unbiased, and has been trained on sufficiently representative data sets; your freedom from pervasive or discriminatory surveillance and monitoring in your home, community, and workplace; and your right to meaningful recourse if the use of an algorithm harms you.

Of course, enumerating the rights is just a first step.

In this excerpt from How Social Science Got Better: Overcoming Bias with More Evidence, Diversity, and Self-Reflection (2021, Oxford University Press), I argue that the social sciences were born to guide our practical ambitions and are improving in their ability to guide decision-makers. But their primary area of application is in multidimensional public policy choices, which inevitably combine the uncoverable patterns of social life with our value choices. Because our theories and pursuits stem from our collective goals, we need to reflect on the history of applied social science, which has often had too much hubris and too little diversity. There is no replacing the complexity of social science or the difficulty of building codified knowledge for use by society. But the social sciences are becoming more relevant, diverse, and reflective, learning from their history and public use.

NOAA’s Climate Program Office today launched a newly redesigned version of Climate.gov, NOAA’s award-winning, flagship website that provides the public with clear, timely, and science-based information about climate. The redesign expands the site’s already significant capacity to connect Americans with the resources they need to understand and plan for climate-related risks.

Just a meter or two down, below the topsoil that nurtures crops, is a little known part of the ecosystem that may be critical to the planet’s climate future. But this deep soil is surprisingly hard to study. It helps to know the right backhoe operator, and even then extracting samples without disturbing their structure or inhabitants is tricky. “The deeper you go, the harder it is,” says Daniel Richter, a soil scientist at Duke University.

Last month, the U.S. National Science Foundation announced funding for a new $19 million research facility, called the Deep Soil Ecotron, that aims to make studying this frontier easier. The initial design for the lab, to be built over the next 5 years at the University of Idaho, calls for 24 richly instrumented soil columns topped with airtight chambers for vegetation. These ecosystems-in-a-lab, or ecounits, will allow researchers to manipulate environmental conditions down to 3 meters. Surprises are assured. “It’s kind of like when people launched the first deep-sea submarine,” says Zachary Kayler, a co–principal investigator (co-PI) and biogeochemist at the University of Idaho. “The possibilities are endless.”

from a rich conversation with Élise Lehoux: Les données de recherche

State government interest in artificial intelligence technology is on the rise, according to experts and a state official who spoke at an event here this week.

Nelson Moe, Virginia’s chief information officer, said he and his team see opportunities for incorporating predictive analytics into government agency workflows, and that AI technology could help with things like detecting waste, fraud and abuse, and supporting decisions in areas ranging from finance to traffic management.

“We want to take it up the value chain, across the board,” Moe said during a conference the National Association of State Chief Information Officers held here this week.

The AI and analytics team will be responsible for managing data, establishing best practice, and avoiding duplication of data analysis efforts across the bank’s four divisions, according to the memo. It will operate in a hub-and-spoke model with central staff at the group level coordinating and interacting with “spoke” teams that sit within UBS’s wealth, asset, investment bank, and Swiss units.

The changes reflect [Ralph] Hamers’ first major steps to enact what has become his signature theme since joining the Swiss bank a year ago, namely using digital technology to cut costs. That will entail job cuts, as the chief executive has indicated he wants to use artificial intelligence to help win wealthy clients amid increased competition.

In mechanobiology, cells change their shape and trajectory as they move across fibrous environments in the human body, constantly tugging or pushing on the fibers and modifying the background environment, which in-turn influences the movement of cells in a perpetual loop.

“This is fundamentally different from mainstream applications in computer vision where changes in the background caused by pedestrians and vehicles are far less accelerated than those possible by the movement of living cells governed by the laws of mechanics and biology,” he said.

To address this challenge, the National Science Foundation has awarded a team of Virginia Tech scientists a $1 million grant to create a new avenue of research in physics-guided machine learning. The project will, for the first time, systematically integrate the mechanics of cell motion available as biological rules and physics-based model outputs to predict the movement of shape-shifting objects in dynamic physical environments.

As principal investigator, Karpatne will team with co-principal investigators Amrinder Nain, associate professor, and Sohan Kale, assistant professor in the Department of Mechanical Engineering, combining his expertise in machine learning with their specialties in cell mechanobiology and computational modeling, respectively.

The expertise that landed Luke Bornn a tenure track faculty position at Harvard was his understanding of movement in time and space. Initially, that focus was on “herding dynamics of animals and climate systems,” he says, but then he stumbled upon NBA player-tracking data.

“It interested me not because I was a huge sports fan,” he notes, “but because it was the richest space-time data I’d ever seen.”

Bornn ultimately left academia and has had a distinguished career in sports as the head of analytics departments for franchises in both European soccer (Serie A’s A.S. Roma) and in basketball (the Sacramento Kings). In his latest venture, Zelus Analytics—which he co-founded alongside Doug Fearing, the former R&D boss of baseball’s Los Angeles Dodgers and Tampa Bay Rays—Bornn will lead the new NBA product launching this season with five franchises locked in and a sixth being finalized.

“If you look around the league, even the teams that are investing most heavily in analytics don’t really have the capabilities or the capacity to deal with the raw tracking data,” says Bornn, who notes it was “too all-consuming” even for his staff of nine with the Kings.

We’ll keep teaching, and doing research, and fighting back. But this is hard. 2/n

The Icahn School of Medicine at Mount Sinai has launched a new department dedicated to advancing artificial intelligence (AI) to transform health care, further positioning the Mount Sinai Health System as a leader in providing patient care through pioneering innovations and technologies. The Department of Artificial Intelligence and Human Health is the first department of its kind within a medical school in the United States.

The department’s mission is to lead the artificial intelligence-driven transformation of health care through innovative research, apply that knowledge to treatment in hospital and clinical settings, and provide personalized care for each patient, which will expand Mount Sinai’s impact on human health across the Health System and around the world.

Albion needed more students for a simple reason: More students equal more money, at least in theory. Without state aid, private colleges are dependent on tuition, room and board to keep their doors open. At Albion, those three categories brought in 58% of the school’s total revenue in the 2018-19 school year.

But schools often find the only way to bring more students on to campus is to give hefty price breaks, which is exactly what happened at Albion.

Armed with discounts, recruiters went into heavily minority areas where the college previously had not recruited. They were forced to look in new areas for students because of a shrinking pool of high school graduates in Michigan and intense competition for them among colleges and universities.

By all accounts, deep neural networks like VGG have way too many parameters and should overfit. But they don’t. Instead, such networks generalize astoundingly well to new data — and until recently, no one knew why. It wasn’t for lack of trying. For example, Naftali Tishby, a computer scientist and neuroscientist at the Hebrew University of Jerusalem who died in August, argued that deep neural networks first fit the training data and then discard irrelevant information (by going through an information bottleneck), which helps them generalize. But others have argued that this doesn’t happen in all types of deep neural networks, and the idea remains controversial.

Now, the mathematical equivalence of kernel machines and idealized neural networks is providing clues to why or how these over-parameterized networks arrive at (or converge to) their solutions. Kernel machines are algorithms that find patterns in data by projecting the data into extremely high dimensions.

It’s going to be very interesting to watch the politics around datasets play out. Effectiveness of this data-centric approach suggests people are going to try and (eventually) bottle up all the text on the internet to feed their models.

Official PyTorch implementation of the NeurIPS 2021 paper