Data Science newsletter – October 30, 2019

GROUP CURATION: N/A

Hundreds of millions of people worldwide are at risk of losing their homes as entire cities sink under rising seas over the next three decades, according to researchers.

The findings, published Tuesday in the journal Nature Communications, put nearly three times as many people in coastal areas at risk from flooding than previously thought, and are the result of new advances in elevation modeling technology.

Ceres2030, a global effort led by International Programs in the College of Agriculture and Life Sciences (IP-CALS), the International Food Policy Research Institute and the International Institute of Sustainable Development, is employing machine learning, librarian expertise and cutting-edge research analysis to use existing knowledge to help solve these and other challenges – all aimed at eliminating hunger by 2030.

“We have an opportunity to achieve higher food security, create a safety blanket to cope with climate shocks and improve overall livelihoods,” said Maricelis Acevedo.

Microsoft has invested in a startup that uses facial recognition to surveil Palestinians throughout the West Bank, in spite of the tech giant’s public pledge to avoid using the technology if it encroaches on democratic freedoms.

AnyVision, which is headquartered in Israel but has offices in the United States, the United Kingdom and Singapore, sells an “advanced tactical surveillance” software system, Better Tomorrow. It lets customers identify individuals and objects in any live camera feed, such as a security camera or a smartphone, and then track targets as they move between different feeds.

According to five sources familiar with the matter, AnyVision’s technology powers a secret military surveillance project throughout the West Bank. One source said the project is nicknamed “Google Ayosh,” where “Ayosh” refers to the occupied Palestinian territories and “Google” denotes the technology’s ability to search for people.

Those who want to reform the criminal justice system are placing their bets on a new wave of prosecutors who have been voted in around the country. Elect a district attorney who will pursue fewer cases, the reasoning goes, and fewer people will be drawn into jails and prisons.

In 2016, Kim Foxx unseated an incumbent in Cook County, Illinois, vowing to transform the nation’s second-largest local prosecutor’s office and to bring more accountability to shootings by police while also reducing unnecessary prosecutions for low-level, non-violent crimes.

One year into her term, Foxx did something no other state’s attorney had ever done: she released six years of data outlining what happened in every felony brought to her office, offering an unprecedented view into the decision-making of prosecutors and its impact.

Once a strong proponent of self-driving cars, Wozniak says he doesn’t think either will live up to expectations.
“A couple of years ago I gave up” on believing in autonomous vehicles and many of the promises of AI, Wozniak said.

Decisions made by human-AI teams (
e.g., AI-advised humans) are increasingly common in high-stakes domains such as healthcare, criminal justice, and finance. Achieving high
team performance depends on more than just the accuracy of the AI system: Since the human and the AI may have different expertise, the highest team performance is often reached
when they both know how and when to complement one another. We focus on a factor that is crucial to supporting such complementary: the human’s mental model of the AI capabilities, specifically the AI system’s
error boundary (i.e. knowing “When does the AI err?”). Awareness of this lets the human decide when to accept or override the AI’s recommendation. We highlight two key properties of an AI’s error boundary, parsimony and stochasticity, and a property of the task, dimensionality. We show experimentally how these properties affect humans’ mental models of AI capabilities and the resulting team performance. We connect our evaluations to related work and propose goals, beyond accuracy, that merit consideration during model selection and optimization to improve overall human-AI team performance.

Critics of recent advances in artificial intelligence complain that although these advances have produced remarkable improvements in AI systems, these systems still do not exhibit “real”, “true”, or “genuine” understanding. The use of words like “real”, “true”, and “genuine” imply that “understanding” is binary. A system either exhibits “genuine” understanding or it does not. The difficulty with this way of thinking is that human understanding is never complete and perfect. In this article, I argue that “understanding” exists along a continuous spectrum of capabilities. Consider, for example, the concept of “water”. Most people understand many properties of water: it is wet, you can drink it, plants need it, it forms ice if chilled, and so on. But unfortunately, many people do not understand that water is an electrical conductor and, therefore, one should not use a blowdryer in the shower. Nonetheless, we do not say of those people that they lack “real”, “true”, or “genuine” understanding of “water”. Instead, we say that their understanding is incomplete.

We should adopt this same attitude toward assessing our AI systems.

The dense circuit structure of the mammalian brain remains unknown. The roughly 86 billion neurons use extremely densely packed circuit networks for communication. Developments in three-dimensional electron microscopy (3D EM) have made possible the imaging of brain regions rich in synaptic connections, but dense reconstruction of connectomes has been challenging.

Now, a team from the department of connectomics at the Max Planck Institute for Brain Research in Germany have imaged and analyzed a piece of tissue from the cerebral cortex of a four-week-old mouse, obtained via biopsy from the somatosensory cortex, a part of the cortex occupied with the representation and processing of touch.

A kind of toxic debt is embedded in much of the infrastructure that America built during the 20th century. For decades, corporate executives, as well as city, county, state, and federal officials, not to mention voters, have decided against doing the routine maintenance and deeper upgrades to ensure that electrical systems, roads, bridges, dams, and other infrastructure can function properly under a range of conditions. Kicking the can down the road like this is often seen as the profit-maximizing or politically expedient option. But it’s really borrowing against the future, without putting that debt on the books.

In software development, engineers have long noted that taking the easy way out of coding problems builds up what they call “technical debt,” as the tech journalist Quinn Norton has written.

Like other kinds of debt, this debt compounds if you don’t deal with it, and it can distort the true cost of decisions. If you ignore it, the status quo looks cheaper than it is. At least until the off-the-books debt comes to light.

When the chief of the Pentagon’s new Joint Artificial Intelligence Center briefed reporters recently, he made a point of emphasizing the imminently practical – even potentially boring – applications of machine learning to the business of war.

There’s the “predictive maintenance” that AI can bring to Black Hawk helicopters, for example, and “intelligent business automation” likely to lead to exciting boosts in “efficiencies for back office functions,” Lt. Gen. Jack Shanahan said. There are humanitarian pluses, too: AI will help the Defense Department better manage disaster relief.

But for 2020, the JAIC’s “biggest project,” General Shanahan announced, will be what the center has dubbed “AI for maneuver and fires.” In lulling U.S. military parlance, that includes targeting America’s enemies with “accelerated sensor-to-shooter timelines” and “autonomous and swarming systems” of drones – reminders that war does, after all, often involve killing people.

Gina Ciavarra is sitting in a dark room at NYU Langone Health in Manhattan. It’s a reading room, a space for radiologists like her to examine X-ray and MRI scans. The monitors in front of her display grayscale images of a de-identified patient’s knee, and in them she detects one key problem: a torn ACL. “This is definitely abnormal,” Ciavarra explains.

But there’s another evaluation that Ciavarra must make, in addition to scanning the swirls of bone, ligaments, fat, cartilage, and tendons for problems like tears or arthritis. Was this particular knee scan created by artificial intelligence, or did it emerge from an MRI machine the traditional way? “My gut says it’s AI,” she says, without certainty. “It just looks a little blurry.”

Ciavarra and her NYU colleagues were participants in a study that pitted the quality of AI-created scans against traditional ones. By pairing artificial intelligence with MRI machines, computer scientists and radiologists think they can greatly speed up a common type of medical exam—a boon for patients and hospitals alike. That could mean cutting a ten-minute knee scan to five minutes, or an hour-long cardiac scan to half an hour. It could also save hospitals money, and reduce the need to anesthetize pediatric patients who may have trouble holding still.

The study, which NYU is now preparing to submit for academic review, is part of a project between two strange bedfellows: the NYU School of Medicine and Facebook.

Google, given its status as a supposed neutral search engine, has been allergic to widespread crackdowns on others’ content. It thinks of itself as an index to the open web. At the same time, the company is making increased inroads in the commerce space and is increasingly asking brands to invest more in its updated and ever-expanding platform; a systemic problem pertaining to counterfeit products could easily erode trust. While Google’s recourse seems to be to refuse to crack down on counterfeit content in Search unless legal means are sought, this problem could have a big impact on its brand perception if it continues to try and control more parts of the online experience.

A new report from brand protection software provider Incopro highlights how counterfeit items are rampant on Google search results. The company researched the problem by performing a series of online searches about branded items and categorizing whether the results linked to pages hawking fake goods. It used six examples of a branded product — for example: for pharmaceuticals, Incopro searched for the drug Bactrin; for children’s products it searched for a specific baby teether. In all, up to 60% of the results Incopro found were for “websites and other locations that offer consumer products that are either counterfeit or otherwise infringe brand owner rights,” the report said.

Three individuals sitting in separate rooms collaborated to correctly orient a block so that it could fill a gap between other blocks in a video game. Two individuals who acted as “senders” could see the gap and knew whether the block needed to be rotated to fit. The third individual, who served as the “receiver,” was blinded to the correct answer and needed to rely on the instructions sent by the senders.

The two senders were equipped with electroencephalographs (EEGs) that recorded their brain’s electrical activity. Senders were able to see the orientation of the block and decide whether to signal the receiver to rotate it. They focused on a light flashing at a high frequency to convey the instruction to rotate or focused on one flashing at a low frequency to signal not to do so. The differences in the flashing frequencies caused disparate brain responses in the senders, which were captured by the EEGs and sent, via computer interface, to the receiver. A magnetic pulse was delivered to the receiver using a transcranial magnetic stimulation (TMS) device if a sender signaled to rotate. That magnetic pulse caused a flash of light (a phosphene) in the receiver’s visual field as a cue to turn the block. The absence of a signal within a discrete period of time was the instruction not to turn the block.

Upselling customers on high-tech breast cancer screenings is just one way the 3D mammography industry aggressively promotes its product.

A KHN investigation found that manufacturers, hospitals, doctors and some patient advocates have put their marketing muscle ― and millions of dollars ― behind 3D mammograms. The juggernaut has left many women feeling pressured to undergo screenings, which, according to the U.S. Preventive Services Task Force, haven’t been shown to be more effective than traditional mammograms.

“There’s a lot of money to be made,” said Dr. Steven Woloshin, director of the Center for Medicine and Media at The Dartmouth Institute for Health Policy and Clinical Practice, who published a study in January showing that the health care industry spends $30 billion a year on marketing.

Sand on beach may look all the same, but it’s not. Researchers have found that the material has a “sound,” one that can be linked to its home. Find the source, and we can learn more about how it moves around the world.

Boston, MA, and Online November 14, starting at 12 p.m., Harvard T.H. Chan School of Public Health. “This Forum event will convene a dynamic panel of former governors, who will share their unique insights into the challenges of leadership and natural disasters. What does it take to prepare, respond and rebuild? What roles do the public, local and state officials and emergency responders play? What is the intersection between economies and disasters? And what climate change considerations need to be understood?”

“A team at the NYU Tandon School of Engineering’s Visualization and Data Analytics (VIDA) lab, led by Claudio Silva, professor in the department of computer science and engineering, developed a framework called VisFlow, by which those who may not be experts in machine learning can create highly flexible data visualizations from almost any data.”

Authors: Yuxin Ma, Tiankai Xie, Jundong Li, Ross Maciejewski

Much of model-based reinforcement learning involves learning a model of an agent’s world, and training an agent to leverage this model to perform a task more efficiently. While these models are demonstrably useful for agents, every naturally occurring model of the world of which we are aware—e.g., a brain—arose as the byproduct of competing evolutionary pressures for survival, not minimization of a supervised forward-predictive loss via gradient descent. That useful models can arise out of the messy and slow optimization process of evolution suggests that forward-predictive modeling can arise as a side-effect of optimization under the right circumstances. Crucially, this optimization process need not explicitly be a forward-predictive loss. In this work, we introduce a modification to traditional reinforcement learning which we call observational dropout, whereby we limit the agents ability to observe the real environment at each timestep. In doing so, we can coerce an agent into learning a world model to fill in the observation gaps during reinforcement learning. We show that the emerged world model, while not explicitly trained to predict the future, can help the agent learn key skills required to perform well in its environment.