SCI COMMUN### COVID-19 Johnson & Johnson (J&J) last week became the third COVID-19 vaccinemaker to receive emergency use authorization for its product from the U.S. Food and Drug Administration. In contrast to the two-dose vaccines from Moderna and Pfizer authorized earlier, the J&J vaccine—a harmless virus delivering the gene for the spike protein from SARS-CoV-2—proved safe and effective with a single dose. The company intends to deliver 20 million doses to the United States this month, 80 million more by the end of June, and more than 1 billion doses worldwide this year. A placebo-controlled trial that took place in eight countries and involved more than 43,000 participants found that the single shot had 66% efficacy against moderate to severe COVID-19 after 28 days and 85% protection against severe disease. This is below the approximately 95% efficacy against mild disease achieved by the Pfizer and Moderna vaccines, which produce the spike protein using messenger RNA (mRNA)—but the J&J trial included locations in South Africa and Brazil where SARS-CoV-2 variants that may escape vaccine-induced antibodies are now common. (The mRNA vaccine results came before their spread.) No one who received the J&J vaccine in any country was hospitalized or died from COVID-19. The White House also brokered a deal with Merck, a major vaccine producer that dropped its own COVID-19 candidates because of poor performance, to help make the J&J product. ### Archaeology A detailed excavation in what was once a Roman port city has helped archaeologists identify what may be the oldest known pet cemetery. The remains of nearly 600 cats and dogs had been laid in prepared pits and covered with pieces of pottery and textiles, and some wore collars and other adornments. Researchers discovered the graveyard in 2011 outside the ancient city of Berenice, which today lies in Egypt. Features of some skeletons indicated the animals had lived with debilitating injuries and illnesses and survived into old age, indicating the animals were cared for, the researchers reported recently in World Archaeology . ### Art and science “I'm the first author, you're just et al. ,” raps this year's winner of Science 's annual “Dance Your Ph.D.,” a contest that challenges scientists to explain their research through dance. The first place video by Jakub Kubečka, a doctoral student at the University of Helsinki, features an original rap song and choreography, performed by him and two friends (above), explaining the search for atmospheric molecular clusters—groups of atoms that stick together and encourage water vapor to condense into clouds. Kubečka beat out 39 competitors for the $2000 top prize, sponsored by the artificial intelligence firm Primer. The winning entry is at . ### COVID-19 The U.S. National Institutes of Health (NIH) last week announced a multipronged research effort to better understand and treat Long COVID, in which people suffer lingering effects after infection by the pandemic coronavirus. Symptoms include lung problems, heart abnormalities, and enduring fatigue. In December 2020, Congress gave NIH $1.15 billion over 4 years to study the perplexing condition. The agency is inviting applications for research on its natural history, prevalence, and underlying biology and plans an expansive biorepository for samples from volunteers. Last week, the World Health Organization released a policy document estimating 10% of patients remain unwell 12 weeks after being infected. Explanations for the lasting effects have been elusive ( Science , 7 August 2020, p. [614][1]). ### Climate change Countries are drastically lagging in the fossil fuel cuts needed to reach the goals of the Paris climate agreement, a new analysis suggests. Globally, emissions were up by an average of 0.21 billion tons of carbon dioxide (CO2) per year from 2016 to 2019 compared with 2011–15, the Global Carbon Project reported this week in Nature Climate Change . Although 64 countries—most of them wealthy ones that have contributed the most to climate change—cut their CO2 emissions by a collective 0.16 billion tons per year during this time, their reductions must increase 10-fold, to some 1 billion tons annually, to meet the Paris goal of limiting global warming to 2°C. Although the pandemic caused a 7% drop in emissions in 2020, the report says, evidence from previous economic crises suggests emissions will rebound to previous levels unless recovery plans aggressively push decarbonization. ### Immigration U.S. President Joe Biden last week ended a policy, imposed last year by then-President Donald Trump, that had barred most noncitizens not already in the United States from seeking permanent residency and work permits, or green cards. Trump had said issuing new green cards didn't make sense given unemployment caused by the COVID-19 pandemic. But industry groups had challenged the policy, in part because they said it prevented companies from hiring needed scientists and skilled technical workers. In revoking the ban, Biden said it had harmed U.S. businesses “that utilize talent from around the world.” A Trump ban on temporary work permits remains in place but is set to expire on 31 March. ### Climate policy U.S. President Joe Biden's administration last week raised the government's benchmark for the “social cost of carbon,” the estimate it uses in cost-benefit analyses of regulations and other policies to represent the burden that global warming places on present and future generations. The figure will rise to $51 per ton on an interim basis; former President Donald Trump's administration had set it as low as $1. The revised standard restores the level set under former President Barack Obama, adjusted for inflation. The Biden administration may further increase the figure in an update due in January 2022 to reflect increased damages from heat waves and other disasters made worse by global warming. ### Publishing A study of more than 5000 biomedical journals found a pattern of apparent favoritism: In 206 journals, a single author was responsible for between 11% and 40% of the papers published between 2015 and 2019. Of 100 of these “nepotistic” journals given closer scrutiny, the prolific author was the editor-in-chief for about one-quarter and on the editorial board for more than 60%. Prolific authors also enjoyed faster peer reviews, according to a preprint of the study posted last month on the bioRxiv server. A research team did the analysis after scrutinizing publications by microbiologist Didier Raoult of Aix-Marseilles University, who has promoted hydroxychloroquine as a COVID-19 treatment, although most other studies have found no evidence of benefit. Raoult, who now faces disciplinary action by a French medical regulator, appears as an author on one-third of the 728 papers at the journal New Microbes and New Infections , where some of his collaborators serve as editors. ### Science and art An artificial intelligence (AI) program for the first time has written a play, which was staged by actors in Prague's Švanda Theater and premiered online last week. The script, depicting a robot's journey trying to understand humans, was generated by a widely available AI system called GPT-2. Researchers at Charles University helped it start to write the play by feeding it two sentences of dialogue about human experiences, and the software generated more, using related information drawn from the internet. Dramatist David Košt'ák, who tweaked about 10% of the resulting script to ensure it followed a coherent storyline, called its style “abstract.” But AI: When a robot writes a play showcases what the evolving technology can now do, specialists say. Judge the 60-minute play for yourself at . ### Conservation The population of monarch butterflies overwintering in Mexico showed another big drop this year. Researchers counted 2.1 hectares of occupied habitat, down 26% from last year and more than 80% from 2 decades ago, the Center for Biological Diversity said. Six hectares is the minimum considered necessary to avoid a risk of extinction. DINO TRACKS IN PERIL Ongoing mining threatens to destroy China's largest site of dinosaur tracks, researchers reported online on 27 February in Geoscience Frontiers . In 1994, the first of the tracks, 145 million to 120 million years old, were uncovered in a copper mine in China's southwestern Sichuan province. Paleontologists have identified 1928 individual footprints from dozens of individuals, representing ornithopods, theropods, sauropods, and pterosaurs. By 2012, mining had led one of the track-bearing rock faces to collapse, before it was fully studied. NEW MINE HELD UP The Biden administration has delayed a huge Arizona copper mine opposed by archaeologists and Native tribes, who say it will destroy cultural treasures. In 2014 Congress approved giving 970 hectares of federal land at Oak Flat to a mining firm. But this week officials said they want to review an environmental study needed for the transfer. Mine opponents have asked Congress and the courts to kill the project. A WIN AGAINST MALARIA El Salvador last week became the first country in Central America to be certified free of malaria by the World Health Organization. The country became eligible after recording no home-grown cases of the mosquito-borne disease since 2017. Globally, 38 countries and territories have reached this milestone. BIG CANCER FUND A new foundation will provide $250 million for cancer research, one of the largest such gifts ever. Break Through Cancer was financed by a Richmond, Virginia, businessperson whose son died of cancer in 2020. The funding will support research teams drawn from five prominent U.S. university cancer centers that will study cancer types that are difficult to treat and have high mortality rates, including pancreatic and ovarian cancer, glioblastomas, and acute myelogenous leukemia. NO PLACE LIKE HOME Americans value space research aimed at protecting Earth over sending astronauts to other bodies, a survey by Morning Consult says. Sixty-three percent of respondents called monitoring Earth's climate a top or important priority. Just 33% voiced that level of backing for launching astronauts to Mars or the Moon. ### Should peer reviewers be paid? Reviewing journal articles can seem a thankless task. Scientists do the work for free, even as as journals publish ever more papers and some publishers make sizable profits. Even before the COVID-19 pandemic led to a blizzard of submissions, journal editors were reporting that “reviewer fatigue” was making it harder to find volunteers. At the Researcher to Reader conference on scholarly publishing last week, two teams debated a provocative question: Should peer reviewers be paid? Here are some of their arguments. (See a fuller version at .) YES: “There is no downward pressure on the endless use of academic labor. And the easiest way to exert that pressure is to value the task not [only] with recognition, but with the traditional way to support skilled labor in every other industry, which is money.” James Heathers, a former research scientist, now chief scientist at a technology startup NO: “A 2018 survey found that only 17% of respondents selected cash or in-kind payment as something that would make them more likely to accept review requests.” (Nearly half said more explicit recognition of reviewing work from their universities or employers would inspire them to do it.) Alison Mudditt, CEO of PLOS, a nonprofit publisher of open-access articles NO: It could cost $3960 per accepted paper to cover the cost of reviewing, if each reviewer was paid $450, each manuscript received 2.2 reviews, and the journal accepted 25% of submissions. “Surely that money would better spent on the research itself and on solving our most pressing global challenges.” Tim Vines, a publishing consultant YES: “What might very well happen is fewer papers get submitted, because the costs go up. … A contract provides much needed certainty around the time frame, the quality, and the predictability of the review received.” Brad Fenwick, senior vice president at Taylor & Francis, a for-profit publisher NO: “It's completely unrealistic to expect that anybody is going to have either the time or the expertise or the scale to be able to manage and monitor hundreds of thousands of additional new contracts across the publishing system. Just not gonna happen.” A.M. [1]: http://www.sciencemag.org/content/369/6504/614

"By far, the greatest danger of Artificial Intelligence is that people conclude too early that they understand it." Today, AI is an indispensable part of our daily lives. To define it scientifically, it is the field of study of "intelligent agents"; any device that can mimic a human task, perceive its environment, and takes actions that maximize its chance of successfully achieving its goals. AI has helped numerous industries develop novel approaches to solve challenging problems. While all this is enthralling, there is a catch, the black box dilemma.

It is believed that each company must embrace the AI revolution from the smallest local businesses to the largest global players, and recognize how artificial intelligence can have the greatest impact on their business. The boundary for mistake is quickly diminishing, as global supply chains increase in complexity. With the rising rivalry in a linked digital environment, optimizing productivity by reducing uncertainties of all sorts becomes even more important. The increase of supersonic speed and efficiency standards among suppliers of all kinds further emphasizes the need for AI Solutions Company to harness Artificial Intelligence skills in both supply chains and logistics. Artificial Intelligence has experienced a long, zigzagging evolution to get to this opinion of application in logistics.

When he was a boy, Mark Solms obsessed over big existential questions. What happens when I die? What makes me who I am? He went on to study neuroscience but soon discovered that neuropsychology had no patience for such open-ended questions about the psyche. So Solms did something unheard of for a budding scientist. He reclaimed Freud as a founding father of neuroscience and launched a new field, neuropsychoanalysis. Solms had one other obstacle in his path. Born in Namibia, where his father worked for a South African diamond mining company, he grew up under apartheid in South Africa. Solms later worked at a hospital in Soweto, where a military occupation tried to clamp down on protesters. "Once you reach the end of your studies, you're required to join the very same army whose victims I was looking after," he told me.

The Biden administration has quietly imposed temporary limits on counterterrorism drone strikes and commando raids outside conventional battlefield zones like Afghanistan and Syria, and it has begun a broad review of whether to tighten Trump-era rules for such operations, according to officials. The military and the C.I.A. must now obtain White House permission to attack terrorism suspects in poorly governed places where there are scant American ground troops, like Somalia and Yemen. Under the Trump administration, they had been allowed to decide for themselves whether circumstances on the ground met certain conditions and an attack was justified. Officials characterized the tighter controls as a stopgap while the Biden administration reviewed how targeting worked -- both on paper and in practice -- under former President Donald J. Trump and developed its own policy and procedures for counterterrorism kill-or-capture operations outside war zones, including how to minimize the risk of civilian casualties. The Biden administration did not announce the new limits.

Tensor decomposition is a fundamentally challenging problem. Even the simplest case of tensor decomposition, the rank-1 approximation in terms of the Least Squares (LS) error, is known to be NP-hard. Here, we show that, if we consider the KL divergence instead of the LS error, we can analytically derive a closed form solution for the rank-1 tensor that minimizes the KL divergence from a given positive tensor. Our key insight is to treat a positive tensor as a probability distribution and formulate the process of rank-1 approximation as a projection onto the set of rank-1 tensors. This enables us to solve rank-1 approximation by convex optimization. We empirically demonstrate that our algorithm is an order of magnitude faster than the existing rank-1 approximation methods and gives better approximation of given tensors, which supports our theoretical finding.

It is important for a broad range of applications, including policy making [136], medical imaging [30], advertisement [22], the development of medical treatments [189], the evaluation of evidence within legal frameworks [183, 218], social science [82, 96, 246], biology [235], and many others. It is also a burgeoning topic in machine learning and artificial intelligence [17, 66, 76, 144, 210, 247, 255], where it has been argued that a consideration for causality is crucial for reasoning about the world. In order to discover causal relations, and thereby gain causal understanding, one may perform interventions and manipulations as part of a randomized experiment. These experiments may not only allow researchers or agents to identify causal relationships, but also to estimate the magnitude of these relationships. Unfortunately, in many cases, it may not be possible to undertake such experiments due to prohibitive cost, ethical concerns, or impracticality.

Training machines to understand natural language and interact with humans is an elusive and essential task in the field of artificial intelligence. In recent years, a diversity of dialogue systems has been designed with the rapid development of deep learning researches, especially the recent pre-trained language models. Among these studies, the fundamental yet challenging part is dialogue comprehension whose role is to teach the machines to read and comprehend the dialogue context before responding. In this paper, we review the previous methods from the perspective of dialogue modeling. We summarize the characteristics and challenges of dialogue comprehension in contrast to plain-text reading comprehension. Then, we discuss three typical patterns of dialogue modeling that are widely-used in dialogue comprehension tasks such as response selection and conversation question-answering, as well as dialogue-related language modeling techniques to enhance PrLMs in dialogue scenarios. Finally, we highlight the technical advances in recent years and point out the lessons we can learn from the empirical analysis and the prospects towards a new frontier of researches.

Two popular types of machine translation (MT) are phrase-based and neural machine translation systems. Both of these types of systems are composed of multiple complex models or layers. Each of these models and layers learns different linguistic aspects of the source language. However, for some of these models and layers, it is not clear which linguistic phenomena are learned or how this information is learned. For phrase-based MT systems, it is often clear what information is learned by each model, and the question is rather how this information is learned, especially for its phrase reordering model. For neural machine translation systems, the situation is even more complex, since for many cases it is not exactly clear what information is learned and how it is learned. To shed light on what linguistic phenomena are captured by MT systems, we analyze the behavior of important models in both phrase-based and neural MT systems. We consider phrase reordering models from phrase-based MT systems to investigate which words from inside of a phrase have the biggest impact on defining the phrase reordering behavior. Additionally, to contribute to the interpretability of neural MT systems we study the behavior of the attention model, which is a key component in neural MT systems and the closest model in functionality to phrase reordering models in phrase-based systems. The attention model together with the encoder hidden state representations form the main components to encode source side linguistic information in neural MT. To this end, we also analyze the information captured in the encoder hidden state representations of a neural MT system. We investigate the extent to which syntactic and lexical-semantic information from the source side is captured by hidden state representations of different neural MT architectures.

The next section talks about the intersection over union (IoU) which is how an object detection generates the prediction scores.