Data Science newsletter – October 28, 2019

GROUP CURATION: N/A

The Age of AI is coming, and fast, and there is plenty to be concerned about. But the Terminator isn’t one of them.

The real threat? Most of the world, including the United States, is unprepared to reap many of the economic and societal benefits offered by AI or mitigate the inevitable risks. Getting there will take decades. Yet AI applications are advancing faster than our policies or institutions at a time in which science and technology are being under-funded, undersupported and even challenged.

It’s a national emergency in the making.

The National Science Foundation is investing in next-generation infrastructure to support explorations of the cosmos using massive datasets from cutting edge facilities, bolstering a 21st-century global approach to astronomy known as multi-messenger astrophysics. To realize the ambitions of this new field, the NSF has awarded $2.8 million to the University of Wisconsin-Milwaukee (UWM) and eight collaborating organizations to further develop their concept for a Scalable Cyberinfrastructure Institute for Multi-Messenger Astrophysics (SCIMMA).

Multi-messenger astrophysics combines observations of light, gravitational waves and particles to understand some of the most extreme events in the universe. For example, the observation of gravitational waves and light from the collision of two neutron stars in 2017 helped explain the origin of heavy elements, allowed an independent measurement of the expansion of the universe and confirmed the association between neutron-star mergers and gamma-ray bursts.

Honeywell is jumping into robotics — with Pittsburgh as its hub.

The Fortune 100 company is creating Honeywell Robotics, an advanced technology center tasked with developing artificial intelligence, machine learning, computer vision, and advanced robotics for use mainly in warehouses.

Honeywell Robotics will be based at 3 Crossings in the Strip District. The company leased 25,000 square feet of space formerly occupied by Robert Bosch in the development last spring, but did not say why.

A group of all-star computer scientists is behind OctoML, a new University of Washington spinout that aims to help companies deploy machine learning models on various hardware configurations.

The startup launched in July and today announced a $3.9 million seed round led by Madrona Venture Group, with participation from Amplify Partners.
OctoML CEO Luis Ceze. (Madrona Photo)

OctoML is led by the creators of Apache TVM, an open source “deep learning compiler stack” that started as a research project at the UW’s Paul G. Allen School of Computer Science & Engineering a few years ago. It has attracted a thriving community of users including tech giants such as Amazon and Facebook that want to optimize and automate their deep learning models for IoT/edge or cloud deployment on an increasing number of platforms such as phones, cars, health devices, and other use cases.

A new study shows that it may be possible to decrease evaluation bias against female professors by making students aware of the problem.

Researchers tested a method for reducing bias by including a short message about bias in the evaluation forms. This “bias reminder” message explained that evaluations are influenced by students’ unconscious biases about race and gender of instructor. The bias reminder then suggested that students should resist stereotypes about instructors and instead focus on the content of the course.

Cities need to plan ahead for automated vehicles that can drop passengers in urban areas then travel long distances to find free lots or street spaces.

California Gov. Gavin Newsom today (Thursday, Oct. 17) announced the debut of the nation’s first statewide earthquake early warning system that will deliver alerts to people’s cellphones through an app developed at the University of California, Berkeley.

The mobile phone app, MyShake, can provide seconds of warning before the ground starts to shake from a nearby quake — enough time to drop, cover and hold on to prevent injury.

The donation will allow the university to train all students, regardless of major, to better use data to make decisions, and understand computer science and artificial intelligence, Mike Galvin said.

“There is no field anymore that doesn’t require an understanding and a capacity … for data analytics, data creativity and data decision-making,” he said.

The university has grown its computer science department in recent years, in part through a $7.6 million donation from Chicago tech entrepreneur Chris Gladwin in 2015, shortly after he sold his big-data storage company Cleversafe to IBM.

Banks are spending lots on A.I.—and they want to use it to manipulate you into spending more money.

Patients with chronic pain commonly believe their pain is related to the weather. Scientific evidence to support their beliefs is inconclusive, in part due to difficulties in getting a large dataset of patients frequently recording their pain symptoms during a variety of weather conditions. Smartphones allow the opportunity to collect data to overcome these difficulties. Our study Cloudy with a Chance of Pain analysed daily data from 2658 patients collected over a 15-month period. The analysis demonstrated significant yet modest relationships between pain and relative humidity, pressure and wind speed, with correlations remaining even when accounting for mood and physical activity. This research highlights how citizen-science experiments can collect large datasets on real-world populations to address long-standing health questions. These results will act as a starting point for a future system for patients to better manage their health through pain forecasts. [full text]

Care for some of the sickest Americans is decided in part by algorithm. New research shows that software guiding care for tens of millions of people systematically privileges white patients over black patients. Analysis of records from a major US hospital revealed that the algorithm used effectively let whites cut in line for special programs for patients with complex, chronic conditions such as diabetes or kidney problems.

The hospital, which the researchers didn’t identify but described as a “large academic hospital,” was one of many US health providers that employ algorithms to identify primary care patients with the most complex health needs. Such software is often tapped to recommend people for programs that offer extra support—including dedicated appointments and nursing teams—to people with a tangle of chronic conditions.

Conventional wisdom in the laboratory sciences suggests that PhD students develop their research skills as a function of mentorship from their faculty advisors (i.e., principal investigators; PIs). However, no prior research has identified empirically a relationship between specific PI practices and the development of research skills. Here we show that PIs’ laboratory and mentoring activities do not significantly predict students’ skill development trajectories, but the engagement of postdocs and senior graduate students in laboratory interactions do. These findings support the practice of “cascading mentorship” as differentially effective and identify a critical but previously unrecognized role for postdocs in the graduate training process. They also illustrate both the importance and the feasibility of identifying evidence-based practices in graduate education.

The five largest U.S. corporations—Alphabet, Amazon, Apple, Facebook, and Microsoft—are all tech companies with combined market capitalization of over four trillion dollars. Tech is often called “Big Tech” these days. Furthermore, a small number of corporations have come to dominate the IT industry, as within each industry segment one corporation often dominates.

The phenomenon whereby corporate dominance seems to be entrenched is often referred to as “winner takes all.” In the context of tech, such a phenomenon can be partly explained by two “laws:” Metcalfe’s Law asserts that the effect of a communications network is proportional to the square of the number of connected users. This makes Facebook, with over 1.5B daily users, dominant as a social network. Kai-Fu Lee’s Virtuous Cycle asserts “More data begets more users and profit, which begets more usage and data.” This explains, for example, the dominance of the Google search engine. Metcalfe’s Law and the Virtuous Cycle make tech companies into natural monopolies, some claim.

As I argued earlier this year, we need laws and regulations, instead of an ethics outrage, to deal with undesired business models and conduct of tech corporations.

Researchers can now link weather events to emissions – and to the companies responsible. A string of lawsuits is about to give “attribution science” a real-life test.

The deal would make Alphabet a player in the wearable fitness tracking space, competing against the likes of Apple, which recently released a new version of its popular smartwatch. Google licenses its Wear operating system to companies such as Fossil but does not currently make its own smartwatch.
Google has described its hardware strategy as “ambient computing,” meaning users should be able to access its services wherever they are. Buying Fitbit could be a play to make Google services a greater part of customers’ lives and measure up to Apple in the health and fitness space. Google hired former Geisinger Health CEO David Feinberg last year to consolidate its health-care strategy. The company announced several new hardware products earlier in October, including the new Pixel 4 smartphone.

Mountain View, CA December 2-3. “A developer-focused conference that provides a platform for you to exchange knowledge, discuss real-world use cases and solutions, and get hands-on with expert-led, deep-dive training and workshops.​ Along with like-minded developers, data scientists, and innovators, you will experience technologies enabling the Internet of Things (IoT), Artificial Intelligence (AI) and robotics.” [$$$]

Thank you to all our contributors and users for making this possible!

A handful of brief news-y items [about notebooks]…

A fourth of the northern hemisphere is covered in permafrost, and it’s melting, thanks to climate change. While this presents a lot to worry about, a new partnership is ready to help speed up solutions.

The National Science Foundation has awarded $3 million to a four-year collaboration of five scientific institutions – including the Arctic Data Center, which is operated by National Center for Ecological Analysis (NCEAS) – to create the Permafrost Discovery Gateway. This online resource will make data and other scientific information about the changing permafrost accessible and usable by industry, conservation organizations, and people who live and work in the Arctic.

“Condescending language use is caustic; it can bring dialogues to an end and bifurcate communities. Thus, systems for condescension detection could have a large positive impact. A challenge here is that condescension is often impossible to detect from isolated utterances, as it depends on the discourse and social context. To address this, we present TalkDown, a new labeled dataset of condescending linguistic acts in context. We show that extending a language-only model with representations of the discourse improves performance, and we motivate techniques for dealing with the low rates of condescension overall. We also use our model to estimate condescension rates in various online communities and relate these differences to differing community norms.”

“In today’s post, we will discuss an interesting property concerning the trajectory of gradient descent iterates, namely the length of the Gradient Descent curve. Let us assume we want to minimize the function shown in Figure 1 starting from a point A. We deploy gradient descent (GD) to this end, and converge to the minimum point B. The green dashed line represents the direct line joining A and B. GD however does not initially know the location of B. Instead, it performs local updates that direct it towards B, as represented by the red solid line. Consequently, the length of the red path that GD follows is longer than the direct green path. How long will the red path be? We studied this question and arrived at some interesting bounds on the path length (both upper and lower) in various settings (check out our preprint!). We discuss some of these bounds in this post. But first, we motivate why understanding path lengths could be useful for understanding nonconvex optimization.”