Data Science newsletter – October 14, 2020

GROUP CURATION: N/A

India’s economy is in an historic slump, and less economic activity means fewer things being thrown away—and also less demand to make new products from the old. No one had much hope that things would pick up soon.
Khwaja Qureshi, a scrap factory owner whose work has come to a standstill because his workers were forced to flee for their lives during the lockdown, waits for them to return from their villages so his factory can resume work again. Dharavi, Mumbai. 17th July, 2020.

The irony is that Dharavi, which has a population of about 1 million and is probably the most densely packed human settlement on Earth, has largely contained the coronavirus. Thanks to an aggressive response by local officials and the active participation of residents, the slum has gone from what looked like an out-of-control outbreak in April and May to a late-September average of 1.3 cases per day for every 100,000 residents, compared with about 7 per 100,000 in Portugal. That success has made Dharavi an unlikely role model, its methods copied by epidemiologists elsewhere and singled out for praise by the World Health Organization. It’s also a remarkable contrast to the disaster unfolding in the rest of India. The country has recorded more than 6.5 million confirmed cases—putting it on track to soon overtake the U.S.—and over 103,000 deaths.

Dharavi’s economic calamity, however, may be just getting started.

Two projects sponsored by Amazon Web Services (AWS) and the Pittsburgh Health Data Alliance (PHDA) have generated solid use cases for machine learning in clinical care.

Amazon Web Services (AWS) and the Pittsburgh Health Data Alliance (PHDA) collaborated in August 2019 to advance innovation in areas including cancer diagnostics, precision medicine, electronic health records, and medical imaging.

Through the collaboration, researchers from the University of Pittsburgh Medical Center (UPMC), University of Pittsburgh, and Carnegie Mellon University (CMU) received support from Amazon Search Awards on top of existing support from PHDA to use machine learning to dive into various projects.

One of those projects examined machine learning techniques to help experts study breast cancer risk and understand what drives tumor growth.

At the end of September, amidst its usual flurry of fall hardware announcements, Amazon debuted two especially futuristic products within five days of each other. The first is a small autonomous surveillance drone, Ring Always Home Cam, that waits patiently inside a charging dock to eventually rise up and fly around your house, checking whether you left the stove on or investigating potential burglaries. The second is a palm recognition scanner, Amazon One, that the company is piloting at two of its grocery stores in Seattle as a mechanism for faster entry and checkout. Both products aim to make security and authentication more convenient—but for privacy-conscious consumers, they also raise red flags.

Amazon’s latest data-hungry innovations are not launching in a vacuum. The company also owns Ring, whose smart doorbells have had myriad security issues and have been widely criticized for bringing unprecedented surveillance to traditionally semi-private spaces. Meanwhile, the biometric data that Amazon One will collect is particularly sensitive, because unlike a password you can’t simply change it if a hacker steals it or it gets unintentionally exposed. Amazon has a strong record for maintaining the security of its massive cloud infrastructure, but there have been lapses across the sprawling business. The stakes are already phenomenally high; the more data the company holds the more risk it takes on.

“Amazon has a major genomics cloud platform, so maybe they hold your DNA and now they’re going to have your palm as well? Plus all of these devices inside your house. And your purchase history on Prime. That’s a lot of information. That’s a lot of personal information,” says Nina Alli.”

A decade ago at the University of Michigan, a team aware of these issues set out to confront them with the goal of creating smart machines that serve society. They began their work recruiting researchers who were separated in labs across campus in order to bridge the traditional engineering disciplines and include medicine, humanities, social sciences, architecture, design, and law. They created the Robotics Institute.

Located in the College of Engineering, the institute began a graduate program from scratch. This fall, it will train 75 PhDs and over 100 master’s students. Its first class of four PhDs, graduated in 2019, split careers in industry and academia.

Now, in their most visible and pivotal effort, the robotics team will throw open the doors to a new building in 2021 that will bring together labs across campus to become a national centerpiece for robotics research, education, and collaboration.

The Ford Robotics Building, a $75 million and 140,000 square-foot complex on North Campus, has the sleek features to enable groundbreaking work: a three-story indoor fly zone for autonomous aerial vehicles, an outdoor testing playground for walking robots, high-bay garage space for self-driving cars, a lab with one of the most scientifically-advanced floors for rehabilitation and mobility robots such as prosthetics and exoskeletons, a makerspace shop to emphasize hardware along with software, and a Mars yard with imitation red soil to test rovers.

The first San Francisco Bay Area Summer Institute in Computational Social Science (BAY-SICSS) took place virtually this summer from June 15 to July 3, and we’re delighted with the results. We realized the BAY-SICSS vision of bringing together computational social scientists and practitioners for social good—proof it can be done. We hope this summary report provides a template for future BAY-SICSS organizers to continue this important work.

Organized by Jae Yeon Kim, Jaren Haber, and Nick Camp, this kick-off edition of BAY-SICSS was designed to accommodate the public health crisis and resulting shifts in format and research focus. To date, the Summer Institute in Computational Social Science (SICSS)—started in 2017 by Matthew Salganik (Princeton) and Chris Bail (Duke)—has trained more than 700 young scholars, bringing together students, postdoctoral researchers, and junior faculty for 2 weeks of intensive study and interdisciplinary research. BAY-SICSS is unique among the many SICSS partner locations for its focus on bringing together computational social scientists and practitioners for social good.

A Clemson University faculty member will use an award from the National Science Foundation (NSF) to examine middle school students’ data science knowledge and practices through the lens of social issues and gauge students’ sense of empowerment to positively change communities through data science.

Golnaz Arastoopour Irgens, assistant professor of learning sciences in the Clemson University College of Education, said it is a common misconception that data is neutral or free from the influence of social issues or that data has no effect on social issues. She said it is often the case that technology informed by data science, such as search engines or facial recognition software, has been shown to either reinforce discrimination or mischaracterize minority groups.

Because humans design these forms of technology and many more make decisions based on them, a critical eye on how they are developed and how they are utilized becomes necessary. Arastoopour Irgens said it follows that the way we educate students to employ data science and utilize it falls short when social, ethical and political issues are not integrated into that education.

Saqib Shaikh says people who are blind, like himself, typically develop highly organized routines to keep track of their things — putting keys, wallets, canes and other essentials in the same places each time.

But sometimes life gets messy: A child needs help finding a lost stuffed animal, identical garbage bins get moved around on the curb or coats get jumbled together at a party.

Today, a person using Microsoft’s Seeing AI app can point a phone camera at a scene, such as a conference room table, and hear a description of what’s in the frame: laptops, water bottles, power cords, phones. But it would sometimes also be useful for the machine learning algorithms powering the app to recognize objects that are specific to that individual person, said Shaikh, a Microsoft engineer whose team invented Seeing AI.

Until recently, there hasn’t been enough relevant data to train machine learning algorithms to tackle this kind of personalized object recognition for people with vision disabilities. That’s why City, University of London, a Microsoft AI for Accessibility grantee, has launched the Object Recognition for Blind Image Training (ORBIT) research project to create a public dataset from scratch, using videos submitted by people who are blind or have low vision

For regulators like the FDA, the immediate challenge is bringing together experts to craft health data frameworks and policies that will guide the rest of the industry.

“You can talk about the digital platforms, the Digital Health Center of Excellence, … the data you need to look at, but at the end of the day you need the people with the expertise,” FDA Commissioner Dr. Stephen Hahn said today during a HLTH virtual keynote. “So we’re using our hiring authorities and we’re searching for the expertise, and we’ve made a number of terrific personnel hires to help us in this effort.

“One of the things that I said was a priority for the agency when I first came there was unleashing the power of data – this is a part of that. It’s not just about those data; it’s about the people who can help us synthesize, gather and interpret those data to make better and more informed decisions.”

The new gift will enable the university to establish a new school of innovation and entrepreneurship, develop new areas for research, provide numerous scholarships and fellowships for students, and bolster the Rolla region’s economy. … A new, independent, university-affiliated research and development entity modeled after such independent, university-affiliated organizations as the Lawrence Livermore National Laboratory in Livermore, California, and the Johns Hopkins Applied Physics Laboratory in Laurel, Maryland. Four new research centers – focused on infrastructure, advanced manufacturing, artificial intelligence and autonomous systems, and environmental and resource sustainability – will stand at the heart of this entity, which will be supported by research space in the Rolla area. The new entity will serve as the university’s node for partnerships with industry, public and private research foundations, and governmental agencies to stimulate business innovation, provide corporate research and development, and develop prototypes for new products.

Analyst expectations of firms’ earnings are on average biased upwards, and that bias varies over time and stocks, according to new research by experts at Wharton and elsewhere. They have developed a machine-learning model to generate “a statistically optimal and unbiased benchmark” for earnings expectations, which is detailed in a new paper titled, “Man vs. Machine Learning: The Term Structure of Earnings Expectations and Conditional Biases.” According to the paper, the model has the potential to deliver profitable trading strategies: to buy low and sell high. When analyst expectations are too pessimistic, investors should buy the stock. When analyst expectations are excessively optimistic, investors can sell their holdings or short stocks as price declines are forecasted.

“[With the machine-learning model], we can predict how the prices of the stocks will behave based on whether or not the analyst forecast is too optimistic or too pessimistic,” said Wharton finance professor Jules H. van Binsbergen, who is one of the paper’s authors. His co-authors are Xiao Han, a doctoral student at the University of Edinburgh Business School; and Alejandro Lopez-Lira, a finance professor at the BI Norwegian Business School.

The researchers found that the biases of analysts increase “in the forecast horizon,” or in the period when the earnings announcement date is not anytime soon. However, on average, analysts revise their expectations downwards as the date of the earnings announcement approaches.

A seemingly small one-to-two degree change in the global climate can dramatically alter weather-related hazards. Given that such a small change can result in such big impacts, it is important to have the most accurate information possible when studying the impact of climate change. This can be especially challenging in data-sparse areas like Africa, where some of the most dangerous hazards are expected to emerge.

A new data set published in the journal Scientific Data provides high-resolution, daily temperatures from around the globe that could prove valuable in studying human health impacts from heat waves, risks to agriculture, droughts, potential crop failures, and food insecurity.

Data scientists Andrew Verdin and Kathryn Grace of the Minnesota Population Center at the University of Minnesota worked with colleagues at the Climate Hazards Center at the University of California Santa Barbara to produce and validate the data set.

When the pandemic hit, parents scrambled to get enough devices to get their kids for online schooling. But even when they did, not everything went smoothly. Getting multiple people online for hours at a time in a home was one big obstacle; making sure entire communities were able to sign on was another.

Jessica Rosenworcel, the senior Democrat on the Federal Communications Commission, wasn’t surprised. For years, Rosenworcel has talked about the “homework gap,” the term she coined to describe a problem facing communities where kids can’t access the internet because infrastructure is inadequate, their families can’t afford it, or both.

This Fall, the Center for Data Science is pleased to welcome its first incoming cohort of medical track PhD students: Daniel Im, Taro Makino, Boyang Yu, and Weicheng Zhu. Launched this Fall 2020, the CDS and NYU School of Medicine track program aims to train students in the application of core data science skills to healthcare and medicine. CDS Medical Track PhD students follow the core Data Science PhD curriculum and then take additional courses and work with faculty in the School of Medicine in areas including imaging, population health, systems biology/genetics, bioinformatics, neuroscience, and cardiovascular disease. These incoming students are embarking on an important new advance in the partnership between data science methodology and crucial medical research, which has already yielded important breakthroughs in areas such as COVID-19 tracking and breast cancer detection.

Streaming giant Netflix has managed to weave its algorithms to predict almost all of the interactions on the site. In fact, it very nearly governs almost everything the user sees when they log on to watch a program, including (but not limited to): page construction, genre rows, the trending videos a user may see, the order they see the videos in, and the image shown from the content.

While the team does have human support — a team of 40 freelancers who hand tag the content and more than 800 engineers, according to a 2013 WIRED article — the algorithms do a lot of heavy lifting when it comes to recommendations.

Algorithms are so vital to the site Netflix even reportedly collects data on how users click and browse to better serve their members. Xavier Amatriain told the tech publication, “We know what you played, searched for, or rated, as well as the time, date and device. We even track user interaction such as browsing or scrolling behavior. All that data is fed into several algorithms, each optimized for a different purpose.”

Artificial intelligence carries the promise of making industry more efficient and our lives easier. With that promise, however, also comes the fear of job replacement, hollowing out of the middle class, increased income inequality, and overall dissatisfaction. According to the quarterly CNBC/SurveyMonkey Workplace Happiness survey from October last year, 37% of workers between the ages of 18 and 24 are worried about AI eliminating their jobs.

But a recent study from two researchers affiliated with the Stanford Institute for Human-Centered Artificial Intelligence (HAI) challenged this public perception about AI’s impact on social welfare. The study found a relationship between AI-related jobs and increases in economic growth, which in return improved the well-being of the society.

Recall that in Part 1 we created two recommendation engine models on top of our data: a matrix factorization model and a deep one. To do so, we framed the recommendation system as a rating prediction machine learning problem. Note: If you haven’t read Part 1, this won’t make much sense so head over here to catch up.

The embeddings are trainable layers, which means that throughout the training, we learn a new dense representation of users and beers. In this part, we will try to explore these spaces.