Large width limits have been a recent focus of deep learning research: modulo computational practicalities, do wider networks outperform narrower ones? Answering this question has been challenging, as conventional networks gain representational power with width, potentially masking any negative effects. Our analysis in this paper decouples capacity and width via the generalization of neural networks to Deep Gaussian Processes (Deep GP), a class of hierarchical models that subsume neural nets. In doing so, we aim to understand how width affects standard neural networks once they have sufficient capacity for a given modeling task. Our theoretical and empirical results on Deep GP suggest that large width is generally detrimental to hierarchical models. Surprisingly, we prove that even nonparametric Deep GP converge to Gaussian processes, effectively becoming shallower without any increase in representational power. The posterior, which corresponds to a mixture of data-adaptable basis functions, becomes less data-dependent with width. Our tail analysis demonstrates that width and depth have opposite effects: depth accentuates a model's non-Gaussianity, while width makes models increasingly Gaussian. We find there is a "sweet spot" that maximizes test set performance before the limiting GP behavior prevents adaptability, occurring at width = 1 or width = 2 for nonparametric Deep GP. These results make strong predictions about the same phenomenon in conventional neural networks: we show empirically that many neural network architectures need 10 - 500 hidden units for sufficient capacity - depending on the dataset - but further width degrades test performance.

SCI COMMUN### COVID-19 Despite past safety concerns, the Brazilian Health Regulatory Agency (Anvisa) this week decided to allow the importation of 928,000 doses of Sputnik V, the Russian-made vaccine against the pandemic coronavirus. Brazil has one of the world's highest burdens of COVID-19, but only about 15% of its population has received a first dose of vaccine. In April, Anvisa had refused to allow the vaccine into the country, citing allegations that Sputnik V contained adenoviruses that could replicate and harm vaccinated people. But a Brazilian law enacted in March allows the country under certain conditions to selectively import vaccines that Anvisa has not yet authorized for emergency use. The agency will require the batches of vaccine to undergo a safety review by a Brazilian government lab. Anvisa lifted the import ban after pressure from 14 governors who had already made agreements to buy more than 67 million doses of Sputnik V, which more than 60 countries have approved for emergency use. > “It shows we are still fully on the wrong track.” > > Climate scientist Pieter Tans of the U.S. National Oceanic and Atmospheric Administration, in The Washington Post, about May's atmospheric carbon dioxide reading of 419 parts per million, the highest in 63 years of modern recording despite pandemic lockdowns. ### Marine ecology Just 6 months into this year, an alarming 761 manatees have died on Florida's east coast—about 10% of the state's population of this unique vegetarian marine mammal. Most of this year's deaths, which already total more than all in 2020, occurred in the Indian River Lagoon, where about 2000 of these subtropical goliaths typically winter, basking in warm water discharged by a power plant. But increased concentrations of nutrient pollution have triggered algal blooms that block sunlight, decreasing the amount of seagrass, the manatees' main food there. They chose the warm water “even though they starved,” says Martine deWit, a veterinarian with the Florida Fish and Wildlife Conservation Commission. In 2017, the administration of then-President Donald Trump downgraded the species from “endangered” to “threatened” under the Endangered Species Act, citing its growing population. But conservationists have called the move premature. ### Racial justice The National Football League (NFL) said last week that in awarding players compensation for brain injuries under a 2013 legal settlement, it will drop a practice that critics have assailed as racist. Under the “race-norming” policy, a scoring algorithm for dementia that physicians employed in assessing players assumed that Black men started their careers with cognitive skills inferior to those of their white counterparts, making it harder for them to show the same amount of injury-induced cognitive decline as white players and to qualify for monetary awards. A majority of the NFL's roughly 20,000 retirees are Black. In a statement, the NFL noted that race-norming has been used for decades by neuropsychologists, who compare patients' scores with averages for their age, gender, education, and race. The NFL says no “off-the-shelf” alternative exists, so it is convening a panel of eight neuropsychologists, three of them Black, to develop a new algorithm. It will be applied going forward and also retrospectively for Black players who would have received an award had they been white. Previously, the NFL appealed some Black players' claims if their cognitive scores had not been adjusted for race. To date, more than 2000 former players have filed for awards, but fewer than 600 have received them. ### Infectious diseases ![Figure][1] CREDITS: (GRAPHIC) N. DESAI/ SCIENCE ; (DATA) UNAIDS The world has made great progress against AIDS, but ambitious targets have been missed, says an analysis issued last week on the 40th anniversary of the emergence of the disease. The report by the Joint United Nations Programme on HIV/AIDS (UNAIDS) notes that 27.4 million of the 37.6 million people now living with HIV are receiving lifesaving antiretroviral treatment. That's a tripling since 2010, but it falls short of a UNAIDS target, set in 2015, of 30 million in treatment by 2020. Because HIV-infected people who receive antiretrovirals rarely transmit the virus, hitting the treatment target would have averted 3.2 million infections and 1 million deaths over the past 5 years, the report says. And, it says, in 2020 the coronavirus pandemic disrupted treatments and supplies of antiretrovirals, with many countries reporting dips in new diagnoses. ### Publishing Days after publishing a letter alleging Israel's actions had threatened the health of Palestinians, The Lancet removed it from its website, fearing that supporters of Israel would boycott the journal, three of the letter's co-authors asserted last week. The letter, published in March 2020, is still accessible in the ScienceDirect database operated by The Lancet 's owner, publishing giant Elsevier. It argued that Palestinians in the Gaza Strip were ill prepared to confront the COVID-19 pandemic because Israel's security operations had damaged Gaza's public health system. In a commentary last week in The BMJ , three authors—all of whom have worked with Palestinian aid organizations, and two of whom are physicians—said The Lancet 's editor-in-chief, Richard Horton, told them last year that a similar letter it published in 2014 had drawn boycott threats and taken a “traumatic” personal toll on its employees. The prestigious medical journal also published a letter in September 2020 that criticizes the removed letter; it remains on The Lancet 's website. The authors of the March 2020 letter praised other Lancet articles that focused on poor health conditions in Gaza. But they also complained of a double standard and called the letter's removal censorship and “a dangerous new precedent.” The Lancet did not respond before Science 's deadline to a request for comment. ### Public health A strategy for fighting dengue fever using bacteria-armed mosquitoes has passed its most rigorous test yet: a randomized controlled trial in Indonesia. Infecting Aedes aegypti mosquitoes with the bacterium Wolbachia pipientis renders them resistant to infection with the dengue virus and less likely to spread it to people. In previous studies, areas where Wolbachia -infected mosquitoes were released reported fewer cases of dengue than nearby untreated areas. In the new trial, conducted by the nonprofit World Mosquito Program, researchers divided a 26-kilometer area in Yogyakarta into 24 clusters and set out containers of Wolbachia -carrying mosquito eggs in 12 randomly selected clusters. Of people visiting primary care clinics with a fever, 2.3% of those living in the treated clusters tested positive for dengue virus, versus 9.4% of those from control areas—a 77% reduction in infections, the team reported this week in The New England Journal of Medicine . Researchers expect the bacterium will continue to reduce dengue incidence and may even eliminate it in the area: Infected insects pass Wolbachia to offspring, and it remains prevalent among the city's wild mosquitoes more than 3 years after the last egg release. ### Biodiversity An automated system that integrates robotics with machine learning, imaging, and a cutting-edge gene sequencer promises to help speed up the discovery of unknown species of insects, which make up an estimated 90% of all animal species yet to be cataloged. Scientists routinely collect thousands of animals in the field, then face long hours in the lab to identify the specimens. The new technology, called DiversityScanner, plucks individual insects from trays and compares their legs, antennae, and other features to known specimens to classify the insect into one of 14 types. An Oxford Nanopore Technologies sequencer then produces a species-identifying piece of DNA called a barcode. The data and an image of the insect are added to a database. Scientists still have to name and describe new species. Some researchers call the system, designed to be easy for labs to replicate with open-source technology and easily available parts, a potential game changer. The designers—Rudolf Meier, who is moving to Berlin's Museum of Natural History, and colleagues—described it in two preprints posted on bioRxiv in May. ### Anthropology Several people from Okinawa, Japan's southernmost prefecture, are stepping up pressure on Kyoto University to return the remains of 26 people from Okinawan burial caves and sites that were unearthed almost 100 years ago and taken to what was then Kyoto Imperial University for study. Some of the remains, a small portion of the 200 sets removed, are believed to be from the royal family of the Ryukyu Kingdom, which was absorbed into the Japanese empire in 1872. A 2018 lawsuit by a group of five Okinawans against Kyoto University is still pending; frustrated by its slow pace and the university's refusal to cooperate, plaintiffs held an online briefing last week to make their case to the international press. Holding the remains violates Japanese law and the United Nations Declaration on the Rights of Indigenous Peoples, says Yasukatsu Matsushima, an economist at Ryukoku University who leads the legal challenge. Very little research about the remains has been published, and nothing recently, Matsushima says. Kyoto University “does not consider that the bones were obtained illegally,” the institution wrote in a statement. ### Racial justice A prominent Black chemist withdrew from consideration for a professorship at the University of North Carolina (UNC), Chapel Hill, after its trustees did not endorse giving a tenured position to a high-profile journalist whose reporting on the United States's history of racism has raised controversy. UNC faculty informed its chancellor last week that the chemist, Lisa Jones of the University of Maryland, Baltimore, wrote them that the trustees' decision “does not seem in line with a school that says it is interested in diversity.” Her decision came after UNC proposed in January to give the tenured position in journalism to Nikole Hannah-Jones of The New York Times , winner of a Pulitzer Prize for her essay in the “1619 Project,” a 2019 series about slavery that she conceived. The trustees did not act on the proposal; some questioned her academic qualifications. UNC instead offered Hannah-Jones (who is not related to Lisa Jones) a nontenured, 5-year position. Faculty members protested and accused the trustees of submitting to criticism of Hannah-Jones by conservative voices. ### Astronomy In its first year of observations, from 2018 to 2019, the Canadian Hydrogen Intensity Mapping Experiment (CHIME) radio telescope in British Columbia detected 535 fast radio bursts (FRBs)—powerful flashes of radio waves from deep space—more than three times as many as were previously known, researchers announced this week at a meeting of the American Astronomical Society. The bursts are just milliseconds long, which makes it challenging to pinpoint their source. But after CHIME helped trace a nearby FRB to its source in our Galaxy last year, highly magnetized stellar relics called magnetars have emerged as a prime suspect. CHIME catches many FRBs by sweeping the sky with a wide field of view as Earth rotates. Its deep catalog is already paying off: FRBs that flash more than once last slightly longer and have a narrower range of frequencies than one-offs, supporting the idea that the bursts are produced through different mechanisms. ### Planetary science NASA's InSight lander, a mission to study the interior of Mars, got a sorely needed energy boost last month after the agency dusted off its solar panels with a clever technique akin to sandblasting. After InSight's power declined, NASA tried knocking off the dust by jostling the panels with motors originally used to deploy them—without luck. Passing dust devils have also done nothing to clear off the material. So mission engineers had to get crafty. They had the lander's robotic arm scoop up sand and drizzle it above a panel as the wind swept past at up to 21 kilometers per hour. The falling grains bounced off the panel, picking up and carrying away the smaller dust particles, NASA said last week. Controllers noticed an immediate bump in power and a gain of about 30 watt-hours of energy per sol, or martian day. The extra power could help the lander survive aphelion in July, when Mars is farthest from the Sun, and extend the mission for a third full year of listening for tiny marsquakes. [1]: pending:yes

Xbox has quietly announced that it will be making streaming sticks that would allow gamers to play "on any TV or monitor". The news comes as part of an update the video game giant, owned by Microsoft, published ahead of the E3 conference. Streaming sticks may not be the only way that Xbox is looking to get more games into the hands of players, as it also says that it is "working with global TV manufacturers to embed the Xbox experience directly into internet-connected televisions with no extra hardware required except a controller." Microsoft's head of Xbox Phil Spencer had previously said that a dedicated app for the games console could arrive on Smart TVs and stream games directly to them, and it appears like it is now becoming a reality. A number of changes to Xbox Game Pass were also launched alongside this news: the company is working with telecommunication providers on new purchasing models like Xbox All Access, encouraging payments over time rather than money up-front, is rolling Game Pass Ultimate out in Australia, Brazil, Mexico, and Japan, and is adding cloud gaming support directly into the Xbox app on PC.

Just days before Microsoft's big ol' E3 livestream, executives from the company sat down to talk -- er, read remarks prepared by the communications team -- about the future of Xbox. In a pre-recorded media briefing, Xbox head Phil Spencer, Microsoft CEO Satya Nadella and others bragged about how well Game Pass and Azure are performing, and also dropped some news about the company's cloud gaming and subscription strategies. First, Xbox is working with global TV manufacturers to get Game Pass on smart televisions. Considering a Game Pass Ultimate subscription unlocks cloud capabilities, this feature will allow folks to play Xbox titles with just a controller, no console required. Additionally, Microsoft is officially building a video game streaming stick, as Spencer teased late last year.

Tens of thousands of times a year, a technician places a drop of blood on a slide and peers at it under a microscope, searching for malaria parasites. Making a definitive diagnosis requires the technician to look at up to 300 different fields of view over roughly half an hour. This process is repeated over and over, day after day, on every continent except Antarctica. It's tedious work, but it saves lives. Malaria parasites infect over 200 million people and kill 400,000 every year, mostly children in Africa. Trained and experienced malaria microscopists are rare, however.

Deb Campbell, a senior staff member in the HADR Systems Group, started the session with a discussion of how to accelerate the national and global response to climate change. "Because the timeline is so short and challenges so complex, it is essential to make good, evidence-based decisions on how to get to where we need to go," she said. "We call this approach systems analysis and architecture, and by taking this approach we can create a national climate change resilience roadmap." This roadmap implements more of what we already know how to do, for example utilizing wind and solar energy, and identifies gaps where research and development are needed to reach specific goals. One example is the transition to a fully zero-emission vehicle (ZEV) fleet in the United States in the coming decades; California has already directed that all of the state's new car sales be ZEV by 2035.

International challenges have become the de facto standard for comparative assessment of image analysis algorithms given a specific task. Segmentation is so far the most widely investigated medical image processing task, but the various segmentation challenges have typically been organized in isolation, such that algorithm development was driven by the need to tackle a single specific clinical problem. We hypothesized that a method capable of performing well on multiple tasks will generalize well to a previously unseen task and potentially outperform a custom-designed solution. To investigate the hypothesis, we organized the Medical Segmentation Decathlon (MSD) - a biomedical image analysis challenge, in which algorithms compete in a multitude of both tasks and modalities. The underlying data set was designed to explore the axis of difficulties typically encountered when dealing with medical images, such as small data sets, unbalanced labels, multi-site data and small objects. The MSD challenge confirmed that algorithms with a consistent good performance on a set of tasks preserved their good average performance on a different set of previously unseen tasks. Moreover, by monitoring the MSD winner for two years, we found that this algorithm continued generalizing well to a wide range of other clinical problems, further confirming our hypothesis. Three main conclusions can be drawn from this study: (1) state-of-the-art image segmentation algorithms are mature, accurate, and generalize well when retrained on unseen tasks; (2) consistent algorithmic performance across multiple tasks is a strong surrogate of algorithmic generalizability; (3) the training of accurate AI segmentation models is now commoditized to non AI experts.

Technological advances in information sharing have raised concerns about data protection. Privacy policies contain privacy-related requirements about how the personal data of individuals will be handled by an organization or a software system (e.g., a web service or an app). In Europe, privacy policies are subject to compliance with the General Data Protection Regulation (GDPR). A prerequisite for GDPR compliance checking is to verify whether the content of a privacy policy is complete according to the provisions of GDPR. Incomplete privacy policies might result in large fines on violating organization as well as incomplete privacy-related software specifications. Manual completeness checking is both time-consuming and error-prone. In this paper, we propose AI-based automation for the completeness checking of privacy policies. Through systematic qualitative methods, we first build two artifacts to characterize the privacy-related provisions of GDPR, namely a conceptual model and a set of completeness criteria. Then, we develop an automated solution on top of these artifacts by leveraging a combination of natural language processing and supervised machine learning. Specifically, we identify the GDPR-relevant information content in privacy policies and subsequently check them against the completeness criteria. To evaluate our approach, we collected 234 real privacy policies from the fund industry. Over a set of 48 unseen privacy policies, our approach detected 300 of the total of 334 violations of some completeness criteria correctly, while producing 23 false positives. The approach thus has a precision of 92.9% and recall of 89.8%. Compared to a baseline that applies keyword search only, our approach results in an improvement of 24.5% in precision and 38% in recall.

The EU is facing a backlash over new AI rules that allow for limited use of facial recognition by authorities -- with opponents warning the carveouts could usher in a new age of biometric surveillance. A coalition of digital rights and consumer protection groups across the globe, including Latin America, Africa and Asia are calling for a global ban on biometric recognition technologies that enable mass and discriminatory surveillance by both governments and corporations. In an open letter, 170 signatories in 55 countries argue that the use of technologies like facial recognition in public places goes against human rights and civil liberties. "It shows that organizations, groups, people, activists, technologists around the world who are concerned with human rights, agree to this call," said Daniel Leufer of U.S. digital rights group Access Now, which co-authored the letter. The use of facial recognition technology is becoming widespread.

Scientists at Duke University and the Wildlife Conservation Society (WCS) used a deep-learning algorithm--a form of artificial intelligence--to analyze more than 10,000 drone images of mixed colonies of seabirds in the Malvinas/Falkland Islands off Argentina's coast. The Malvinas/Falklands are home to the world's largest colonies of black-browed albatrosses (Thalassarche melanophris) and second-largest colonies of southern rockhopper penguins (Eudyptes c. chrysocome). Hundreds of thousands of birds breed on the islands in densely interspersed groups. The deep-learning algorithm correctly identified and counted the albatrosses with 97% accuracy and the penguins with 87%. All told, the automated counts were within 5% of human counts about 90% of the time.