SCI COMMUN ### Areas to watch As biomedical scientists continue to battle the deadly pandemic this year to help the world return to normalcy, researchers across the disciplines still aim to hit big milestones or launch new projects despite the challenges brought by COVID-19. European scientists will also have to contend with the aftermath of Brexit. Many U.S. scientists, in contrast, have a more hopeful political outlook, with some likely to play an invigorated role in tackling another global crisis, climate change, after President-elect Joe Biden, who has vowed to make it a top priority, is sworn in this month. In this section, Science 's news staff forecasts areas of research and policy we expect to make headlines this year, from protecting the high seas' biodiversity to probing how ancient humans interacted. ### Global warming Nearly 8 years have passed since the fifth assessment report from the United Nations's Intergovernmental Panel on Climate Change, the famed body of volunteer climate scientists that since 1990 has chronicled humanity's persistent warming of the planet. The sixth installment, crafted by more than 700 scientists and delayed by the pandemic, will come out in sections this year and next, and it's expected to further sharpen the picture of the human impact on climate. Its findings will be bolstered by a new generation of climate models and scenarios, fed by indicators of unabated global change: data showing accelerating sea level rise; rapid ice melt at the poles; and waves of extreme heat, drought, and fire. In November, the world's countries will meet in Scotland's Glasgow for the next U.N. climate summit, where they are expected to increase the ambitions of their pledged cuts in greenhouse gases and agree on a full set of rules to implement the Paris agreement. Among the attendees will be the United States, which President-elect Joe Biden has said will rejoin the pact. ### Public health The launch of immunization campaigns in many countries has raised hopes that the COVID-19 pandemic can be brought to an end. But how exactly it started remains murky, and a World Health Organization international team of 10 scientists will travel to China several times this year as part of an investigation into the pandemic coronavirus' origins—a politically sensitive mission because the United States and China have sparred over who is to blame for the pandemic. The team hopes to discover the virus' closest relatives in bats, where and how it jumped to humans, whether another species acted as an intermediate host, and, most important, how we can prevent other pandemic viruses from emerging. ### National security Restoring U.S.-Chinese scientific collaborations to health could be a test of President-elect Joe Biden's success in negotiating with the Asian superpower on trade, immigration, and security concerns. A new U.S. government–sponsored forum on science, technology, and national security will advise the new administration on how to strike the right balance between openness and preventing the theft of new technology. A new U.S. law provides stiff penalties for federal grant applicants who fail to disclose all sources of funding when submitting a proposal, based on the assumption that greater transparency is the best way to monitor ties with China and other entities posing potential threats to U.S. security. University leaders are hoping Biden will reverse President Donald Trump's restrictive immigration policies and avoid ratcheting up economic and political tensions. But prominent Republicans in Congress are threatening even more punitive steps against China. ### Infectious diseases To complement mass vaccinations against COVID-19, drug companies this year will dash to custom design drugs that block the pandemic coronavirus and treat the disease's symptoms. Even if many people receive one of the new, highly effective vaccines approved in recent weeks by regulators, the virus is expected to remain endemic. In 2020, only the antiviral remdesivir and a handful of other drugs, all originally designed to treat other conditions, showed even limited benefits against the disease. To identify new drug candidates, researchers have deployed artificial intelligence and supercomputers, and more than 590 experimental drugs are in development, according to a leading pharmaceutical industry tracker. For example, researchers have high hopes for compounds that disrupt reproduction of the pandemic virus by inhibiting one of its two proteases. Cocktails of such therapies could tame the virus, an approach used successfully against HIV. The protease inhibitors and other compounds look promising in cell and animal studies. But studies on human volunteers are only getting started, and it could take years to pass safety and efficacy reviews. ### Planetary science Mars's thin air makes it hard to slow a probe to a soft landing. Of the 18 robotic probes sent to the planet's surface in the past 50 years, eight have crashed. This year, two more will attempt a touchdown. On 18 February, NASA's SUV-size rover, Perseverance, will take the plunge, slowed by parachutes and retrorockets on a “sky crane” platform. After landing in Jezero crater, near a fossilized river delta, the rover will collect rock samples for eventual return to Earth. Around the same time, China's Tianwen-1 mission will arrive with an orbiter, a lander platform, and a rover the size of a golf cart. Officials have chosen a landing site not far from Jezero, along the southern edge of Utopia Planitia, a broad plain that may have been repaved by ancient flows of mud. A successful touchdown would be China's first on Mars. ### Microscopy Researchers aim this year to sharpen the resolution of cryo–electron microscopy (cryo-EM), a technique for studying protein structure that may yield new insights into their roles in maintaining human health and causing disease. Another technique, x-ray crystallography, has long been the gold standard for mapping individual atoms within a 3D protein structure. But it only works for proteins that can be packed into crystals. Cryo-EM doesn't require crystals, and its resolution has steadily improved over the past decade. In 2020, it crossed the threshold of atomic resolution as researchers using cryo-EM microscopes equipped with improved electron detectors and software mapped the structure of apoferritin, an iron-binding protein. That protein is unusually rigid, which made it easier to hold steady during cryo-EM mapping. Next, researchers want to image less rigid proteins. Success would be a boon to structural biologists, allowing them to generate highly detailed maps of large proteins and complexes of multiple proteins that cannot be crystallized. ### Astronomy The long wait will soon be over: NASA's much-delayed flagship observatory, the James Webb Space Telescope (JWST), is set to finally take to the skies on 31 October. JWST is the successor to the Hubble Space Telescope, with a 6.5-meter-wide mirror that has six times its predecessor's light-gathering power. The gold-coated, honeycombed mirror will be cooled so it can collect the infrared light of distant objects, red-shifted by the universe's expansion. JWST will be sensitive enough to scrutinize the atmospheres of nearby exoplanets for signs of life and gather the light of the universe's first stars and galaxies. The $8.8 billion spacecraft, which will cost billions more and launch years later than originally planned, recently endured final tests—violent shaking to simulate launch. This month, engineers are unfolding its mirror and unfurling its multilayered Sun shield one last time to check that all is well. By midyear, JWST will be packed up and shipped to French Guiana, where it will be loaded onto a European Ariane 5 rocket. Next stop: deep space. ### Energy The Joint European Torus (JET), the world's largest fusion reactor, will this year embark on a campaign to generate substantial amounts of fusion power. The U.K.-based JET is a tokamak, which uses powerful magnets to constrain a hot plasma so that atomic nuclei crash together and fuse, releasing energy. After an upgrade, JET has a new metallic lining and extra heating power; in this year's trials, it will be fed a potent mix of the hydrogen isotopes deuterium and tritium (D-T)—a fuel rarely used because radioactive tritium needs careful handling and cleanup. In 1997, the last time this fuel mix was used, JET generated 16 megawatts of power for a few seconds, well short of the power consumed to make it happen. The new campaign will initially aim for similar power levels but try to sustain them for longer. That will help in planning for the huge ITER reactor, under construction in France, which has a similar shape and lining. ITER is due to start operations in 2025, but will not begin to use D-T fuel until the mid 2030s. ### Nutrition Help may arrive this year for millions of malnourished children who remain sickly and fail to fully recover even after receiving proper nutrition and treatments for being underfed. Pandemic-related disruptions and job losses are expected to cause the number of such children to skyrocket. One problem for underfed children is disruption of the gut microbiome, which leads to an immature, inefficient digestive system that stunts growth. To repair the microbiomes, health specialists are looking forward to results of a study in Bangladesh that evaluated a low-cost nutritional supplement; it's a mix of easy-to-find ingredients, such as chickpeas, bananas, and soy and peanut flours. In 2019, this team reported that in experiments in mice and pigs, followed by a monthlong pilot study of 60 children, the supplement led to gut repair, as indicated by changes in blood markers. The study did not last long enough to test effects on growth. Since then, these researchers have been comparing the intervention with an existing supplement in a larger, 3-month trial of malnourished children. ### Conservation Few rules protect biodiversity in the two-thirds of the ocean that lies outside nations' sovereign waters. This year, the United Nations is expected to finalize the first treaty specifically intended to change that. The pact is expected to provide a way to designate marine protected areas (MPAs) on the high seas; researchers have been developing a list of candidates and supporting evidence. The draft language also sets minimum standards for environmental impact assessments that nations would be required to conduct before starting commercial activities that might harm marine life. A new, international scientific and technical body, similar to one that manages marine life around Antarctica, would review MPA proposals. The treaty draft also provides for a system to manage genetic sequences taken from marine organisms living in the high seas. ### Archaeology Expect to see studies of ancient humans follow new paths this year, as researchers combine analyses of ancient DNA with other molecular and microbial clues to examine social ties and migrations. Scientists will merge DNA evidence with data from proteins and isotopes, as well as microfossils and pathogens from bones, tooth plaque, and fossilized poop. Such studies this year could help determine which early Celtic family members inherited wealth. They could help identify the homeland of the biblical Philistines and clarify the identities of early Anglo-Saxons and Greeks in Europe, as well as mummies in China and Egypt. ### Public health An obscure $4 billion U.S. government fund that compensates people injured by vaccines is poised to become more tight-fisted this month. Changes proposed by the Trump administration will likely take effect in mid-January, making it more complicated and time consuming for people to win a payout if they sustain shoulder injuries after incorrectly administered injections with flu, tetanus, and other vaccines. The new rules won't affect people who might be injured by COVID-19 vaccines, who would need to apply to a different government program for compensation. But claims that other vaccines caused shoulder injuries have grown on the heels of expanded influenza vaccination and a 2010 paper in which government scientists first described “shoulder injury related to vaccine administration.” ### Biomedicine For more than 3 decades, scientists have dreamed of shrinking tumors by shutting off a protein called KRAS whose growth signals drive many cancer types. KRAS was thought to be impervious to drugs, in part because it offered no obvious pockets that inhibitors could target. But multiple companies have now developed compounds that fit into a groove on some cancer-promoting mutant KRAS proteins and curb their signals. The drugs have shown promising results, first in rodents and then in cancer patients. In December 2020, Amgen asked the Food and Drug Administration to review its KRAS drug, sotorasib, setting the stage for approval this year of the first member of this novel drug class. The drug could first be licensed for use in certain lung cancer patients. Another firm is expected to submit its KRAS drug for approval this year as well.

In this paper we present the first model for directly synthesizing fluent, natural-sounding spoken audio captions for images that does not require natural language text as an intermediate representation or source of supervision. Instead, we connect the image captioning module and the speech synthesis module with a set of discrete, sub-word speech units that are discovered with a self-supervised visual grounding task. We conduct experiments on the Flickr8k spoken caption dataset in addition to a novel corpus of spoken audio captions collected for the popular MSCOCO dataset, demonstrating that our generated captions also capture diverse visual semantics of the images they describe. We investigate several different intermediate speech representations, and empirically find that the representation must satisfy several important properties to serve as drop-in replacements for text.

This story originally appeared on WIRED UK. This is not a year we'll look back on fondly. It began with Australia on fire and ends with more than 1.5 million dead in a pandemic. But there have been bright points in this annus horribilis. While many of us saved lives by hunkering down at home watching Netflix, a communal act of selflessness that shouldn't be soon forgotten, progress was made in science, the environment, and even politics--Biden won!

Games potentially provide a wealth of knowledge about our shared cultural past and the development of human civilisation, but our understanding of early games is incomplete and often based on unreliable reconstructions. This paper describes the Digital Ludeme Project, a five-year research project currently underway that aims to address such issues using modern computational techniques.

We consider a multi-agent Markov strategic interaction over an infinite horizon where agents can be of multiple types. We model the strategic interaction as a mean-field game in the asymptotic limit when the number of agents of each type becomes infinite. Each agent has a private state; the state evolves depending on the distribution of the state of the agents of different types and the action of the agent. Each agent wants to maximize the discounted sum of rewards over the infinite horizon which depends on the state of the agent and the distribution of the state of the leaders and followers. We seek to characterize and compute a stationary multi-type Mean field equilibrium (MMFE) in the above game. We characterize the conditions under which a stationary MMFE exists. Finally, we propose Reinforcement learning (RL) based algorithm using policy gradient approach to find the stationary MMFE when the agents are unaware of the dynamics. We, numerically, evaluate how such kind of interaction can model the cyber attacks among defenders and adversaries, and show how RL based algorithm can converge to an equilibrium.

Do state-of-the-art natural language understanding models care about word order - one of the most important characteristics of a sequence? Not always! We found 75% to 90% of the correct predictions of BERT-based classifiers, trained on many GLUE tasks, remain constant after input words are randomly shuffled. Despite BERT embeddings are famously contextual, the contribution of each individual word to downstream tasks is almost unchanged even after the word's context is shuffled. BERT-based models are able to exploit superficial cues (e.g. the sentiment of keywords in sentiment analysis; or the word-wise similarity between sequence-pair inputs in natural language inference) to make correct decisions when tokens are arranged in random orders. Encouraging classifiers to capture word order information improves the performance on most GLUE tasks, SQuAD 2.0 and out-of-samples. Our work suggests that many GLUE tasks are not challenging machines to understand the meaning of a sentence.

Structured information is an important knowledge source for automatic verification of factual claims. Nevertheless, the majority of existing research into this task has focused on textual data, and the few recent inquiries into structured data have been for the closed-domain setting where appropriate evidence for each claim is assumed to have already been retrieved. In this paper, we investigate verification over structured data in the open-domain setting, introducing a joint reranking-and-verification model which fuses evidence documents in the verification component. Our open-domain model achieves performance comparable to the closed-domain stateof-the-art on the TabFact dataset, and demonstrates performance gains from the inclusion of multiple tables as well as a significant improvement over a heuristic retrieval baseline. Figure 1: Example query to be evaluated against two retrieved tables.

This article presents a novel method to sampling on manifolds based on the Dirichlet distribution. The proposed strategy allows to completely respect the underlying manifold around which data is observed, and to do massive samplings with low computational effort. This can be very helpful, for instance, in neural networks training process, as well as in uncertainty analysis and stochastic optimization. Due to its simplicity and efficiency, we believe that the new method has great potential. Three manifolds (two dimensional ring, Mobius strip and spider geometry) are considered to test the proposed methodology, and then it is employed to an engineering application, related to gas seal coefficients.

Open-domain dialogue agents have vastly improved, but still confidently hallucinate knowledge or express doubt when asked straightforward questions. In this work, we analyze whether state-of-the-art chit-chat models can express metacognition capabilities through their responses: does a verbalized expression of doubt (or confidence) match the likelihood that the model's answer is incorrect (or correct)? We find that these models are poorly calibrated in this sense, yet we show that the representations within the models can be used to accurately predict likelihood of correctness. By incorporating these correctness predictions into the training of a controllable generation model, we obtain a dialogue agent with greatly improved linguistic calibration.

Asymmetric binary classification problems, in which the type I and II errors have unequal severity, are ubiquitous in real-world applications. To handle such asymmetry, researchers have developed the cost-sensitive and Neyman-Pearson paradigms for training classifiers to control the more severe type of classification error, say the type I error. The cost-sensitive paradigm is widely used and has straightforward implementations that do not require sample splitting; however, it demands an explicit specification of the costs of the type I and II errors, and an open question is what specification can guarantee a high-probability control on the population type I error. In contrast, the Neyman-Pearson paradigm can train classifiers to achieve a high-probability control of the population type I error, but it relies on sample splitting that reduces the effective training sample size. Since the two paradigms have complementary strengths, it is reasonable to combine their strengths for classifier construction. In this work, we for the first time study the methodological connections between the two paradigms, and we develop the TUBE-CS algorithm to bridge the two paradigms from the perspective of controlling the population type I error.