Building equitable and inclusive NLP technologies demands consideration of whether and how social attitudes are represented in ML models. In particular, representations encoded in models often inadvertently perpetuate undesirable social biases from the data on which they are trained. In this paper, we present evidence of such undesirable biases towards mentions of disability in two different English language models: toxicity prediction and sentiment analysis. Next, we demonstrate that the neural embeddings that are the critical first step in most NLP pipelines similarly contain undesirable biases towards mentions of disability. We end by highlighting topical biases in the discourse about disability which may contribute to the observed model biases; for instance, gun violence, homelessness, and drug addiction are over-represented in texts discussing mental illness.

The critic Raymond Williams once wrote that every historical period has its own "structure of feeling." How everything seemed in the nineteen-sixties, the way the Victorians understood one another, the chivalry of the Middle Ages, the world view of Tang-dynasty China: each period, Williams thought, had a distinct way of organizing basic human emotions into an overarching cultural system. Each had its own way of experiencing being alive. In mid-March, in a prior age, I spent a week rafting down the Grand Canyon. When I left for the trip, the United States was still beginning to grapple with the reality of the coronavirus pandemic.

Mountain gorillas have been caught on camera as they "sing" during their supper, a behavior that has never before been documented on video. Filmmakers captured the astonishing footage of the primate crooners with a little help from a very special camera: a robotic "spy" designed to look like a young gorilla. The singing apes make their television debut on April 29 in the returning PBS series, "Nature: Spy in the Wild 2." Like its predecessor, which first aired in 2017, the program documents remarkable up-close glimpses of elusive wildlife behavior, seen through the "eyes" of robots that are uncanny lookalikes of the creatures that they film. But this time, the robot animals display an even greater range of realistic behaviors, enabling them to interact with the wildlife that they're spying on. Though human camera operators typically keep a safe distance from wild gorillas, the lifelike animatronic gorilla spy was able to infiltrate a troop and film their daily routines, which included an impromptu suppertime serenade. Footage of the singing gorillas is featured in the first episode of "Spy in the Wild 2" and shows the apes reclining amid dense foliage in a sanctuary in Uganda.

A pack of Silverback Mountain gorillas in Uganda were caught for the first time on camera performing a supper serenade. Filmmakers captured the unique ritual by placing a robotic'spy' that resembles a young gorilla deep in the jungle. The team designed the animatronic machine with realistic eye movements, as wild gorillas communicate with each other through eye contact, and a submissive demeanor with the hopes it would be accepted by the pack. Along with the singing, the footage shows the gorillas screamed a'chorus of appreciation' while eating and provided evidence that they are extremely gassy. A pack of Silverback Mountain gorillas in Uganda were caught for the first time on camera performing a supper serenade.

This paper proposes a simple method to extract from a set of multiple related time series a compressed representation for each time series based on statistics for the entire set of all time series. This is achieved by a hierarchical algorithm that first generates an alphabet of shapelets based on the segmentation of centroids for clustered data, before labels of these shapelets are assigned to the segmentation of each single time series via nearest neighbor search using unconstrained dynamic time warping as distance measure to deal with non-uniform time series lenghts. Thereby, a sequence of labels is assigned for each time series. Completion of the last label sequence permits prediction of individual time series. Proposed method is evaluated on two global COVID-19 datasets, first, for the number of daily net cases (daily new infections minus daily recoveries), and, second, for the number of daily deaths attributed to COVID-19 as of April 27, 2020. The first dataset involves 249 time series for different countries, each of length 96. The second dataset involves 264 time series, each of length 96. Based on detected anomalies in available data a decentralized exit strategy from lockdowns is advocated.

Fall Armyworm is spreading fast across sub-Saharan Africa, devastating crops and farmers' livelihoods. Experts fear the pest could eventually spread to the Middle East and Europe. But a new mobile phone app'Nuru', which uses machine learning and artificial intelligence, offers some hope in tackling the pest problem. Allan Hruska is the Principal Technical Coordinator on the Fall Armyworm response at the Food and Agricultural Organization (FAO).

BOGOTA – About 100 million people living in cities worldwide will likely fall into poverty due to the coronavirus pandemic, urban experts said on Wednesday, calling for mapping tools to identify vulnerable communities and investment focusing on slums. Densely populated cities are at the front line of the contagious outbreak. People living in poverty with little or no running water, sewage systems or health care access have been hit especially hard, said experts at the World Bank, the World Resources Institute (WRI) and other groups studying urban issues. "Within cities we need to focus on those who need help the most, the poor and the vulnerable have been very seriously affected," said Sameh Wahba, global director for the World Bank's urban, disaster risk management, resilience and land global practice. "Our estimate is that there will be possibly upward of a 100 million so-called new poor on account of loses of jobs and livelihoods and income," Wahba told a webinar with members of the media.

A United Arab Emirates (UAE) drone strike on a biscuit factory near the Libyan capital Tripoli on November 18 killed eight civilians and injured 27 others, Human Rights Watch (HRW) said. In a report released on Wednesday, the rights group said the UAE appeared to take little or no action to minimise civilian casualties and called on Emirati authorities to conduct a transparent investigation into the incident. "Since the current armed conflict in Tripoli erupted in April 2019, the UAE has been conducting air and drone strikes to support the Libyan Arab Armed forces, previously known as the Libyan National Army [LNA], one of two major parties to the conflict, some of which have resulted in civilian casualties," HRW said. "All causalities in the November incident were civilian factory workers, including seven Libyans and 28 foreign nationals, all of them men." Human Rights Watch visited the site and found remnants of at least four Blue Arrow-7 (BA-7) laser-guided missiles that were launched by a Wing Loong-II drone.

Following a full-day livestream, we now have a better idea of when and where the next game in Ubisoft's long-running Assassin's Creed franchise will take place. In Assassin's Creed: Valhalla, you'll play an assassin at some point during the Viking Age, which took place between 793 and 1066 CE. Publisher and developer Ubisoft picked the unconventional route of announcing the new game through a Photoshop livestream. An artist slowly and painstakingly built out the image you see above while fans speculated about the setting and classic songs from the series like "Ezio's Family" played in the background. At one point in the stream, more than 50,000 people across Twitch and YouTube tuned in to watch artist Kode Abdo work his craft.

We describe our submission to the Extreme Value Analysis 2019 Data Challenge in which teams were asked to predict extremes of sea surface temperature anomaly within spatio-temporal regions of missing data. We present a computational framework which reconstructs missing data using convolutional deep neural networks. Conditioned on incomplete data, we employ autoencoder-like models as multivariate conditional distributions from which possible reconstructions of the complete dataset are sampled using imputed noise. In order to mitigate bias introduced by any one particular model, a prediction ensemble is constructed to create the final distribution of extremal values. Our method does not rely on expert knowledge in order to accurately reproduce dynamic features of a complex oceanographic system with minimal assumptions. The obtained results promise reusability and generalization to other domains.