We describe our system for SemEval-2020 Task 11 on Detection of Propaganda Techniques in News Articles. We developed ensemble models using RoBERTa-based neural architectures, additional CRF layers, transfer learning between the two subtasks, and advanced post-processing to handle the multi-label nature of the task, the consistency between nested spans, repetitions, and labels from similar spans in training. We achieved sizable improvements over baseline fine-tuned RoBERTa models, and the official evaluation ranked our system 3rd (almost tied with the 2nd) out of 36 teams on the span identification subtask with an F1 score of 0.491, and 2nd (almost tied with the 1st) out of 31 teams on the technique classification subtask with an F1 score of 0.62.

Deep models, while being extremely flexible and accurate, are surprisingly vulnerable to "small, imperceptible" perturbations known as adversarial attacks. While the majority of existing attacks focus on measuring perturbations under the $\ell_p$ metric, Wasserstein distance, which takes geometry in pixel space into account, has long been known to be a suitable metric for measuring image quality and has recently risen as a compelling alternative to the $\ell_p$ metric in adversarial attacks. However, constructing an effective attack under the Wasserstein metric is computationally much more challenging and calls for better optimization algorithms. We address this gap in two ways: (a) we develop an exact yet efficient projection operator to enable a stronger projected gradient attack; (b) we show that the Frank-Wolfe method equipped with a suitable linear minimization oracle works extremely fast under Wasserstein constraints. Our algorithms not only converge faster but also generate much stronger attacks. For instance, we decrease the accuracy of a residual network on CIFAR-10 to $3.4\%$ within a Wasserstein perturbation ball of radius $0.005$, in contrast to $65.6\%$ using the previous Wasserstein attack based on an \emph{approximate} projection operator. Furthermore, employing our stronger attacks in adversarial training significantly improves the robustness of adversarially trained models.

The National Institutes of Health has launched the Medical Imaging and Data Resource Center (MIDRC), an ambitious effort that will harness the powers of artificial intelligence, and medical imaging to fight COVID-19. The multi-institutional collaboration, led by the National Institute of Biomedical Imaging and Bioengineering (NIBIB), part of NIH, will create new tools that physicians can use for early detection and personalized therapies for COVID-19 patients. "This program is particularly exciting because it will give us new ways to rapidly turn scientific findings into practical imaging tools that benefit COVID-19 patients," said Bruce J. Tromberg, PhD, NIBIB Director. "It unites leaders in medical imaging and artificial intelligence from academia, professional societies, industry, and government to take on this important challenge." The features of infected lungs and hearts seen on medical images can help assess disease severity, predict response to treatment, and improve patient outcomes.

Sweeping changes to England's planning system will "cut red tape, but not standards," Housing Secretary Robert Jenrick has said. Under draft new laws, first revealed on Sunday, developers will be granted "automatic" permission to build homes and schools on sites for "growth". It follows Boris Johnson's pledge to "build back better" after coronavirus. But critics warn it could lead to "bad-quality housing" and loss of local control over development. Mr Johnson promised to speed up investment into homes and infrastructure in June to help the UK recover from the economic impact of coronavirus.

Data prepper Tamr Inc. will assist the U.S. Air Force in boosting utilization of its air assets under a five-year contract designed to use machine learning techniques to accelerate the flight certification process for new aircraft configurations. Those configurations include equipping front-line aircraft with new weapons, sensors and defenses such as electronic warfare pods. Tamr said the contract with the Air Force's Seek Eagle Office could be worth as much $60 million. The office based at Eglin Air Force Base, Fla., is responsible for integration new technologies into front-line aircraft. The Air Force office will use Tamr's machine learning platform to organize more than 30 years of aircraft performance studies dispersed across the organization.

PHILADELPHIA - To answer medical questions that can be applied to a wide patient population, machine learning models rely on large, diverse datasets from a variety of institutions. However, health systems and hospitals are often resistant to sharing patient data, due to legal, privacy, and cultural challenges. An emerging technique called federated learning is a solution to this dilemma, according to a study published Tuesday in the journal Scientific Reports, led by senior author Spyridon Bakas, PhD, an instructor of Radiology and Pathology & Laboratory Medicine in the Perelman School of Medicine at the University of Pennsylvania. Federated learning -- an approach first implemented by Google for keyboards' autocorrect functionality -- trains an algorithm across multiple decentralized devices or servers holding local data samples, without exchanging them. While the approach could potentially be used to answer many different medical questions, Penn Medicine researchers have shown that federated learning is successful specifically in the context of brain imaging, by being able to analyze magnetic resonance imaging (MRI) scans of brain tumor patients and distinguish healthy brain tissue from cancerous regions.

The creation of the Global Partnership on Artificial Intelligence (GPAI) reflects the growing interest of states in AI technologies. The initiative, which brings together 14 countries and the European Union, will help participants establish practical cooperation and formulate common approaches to the development and implementation of AI. At the same time, it is a symptom of the growing technological rivalry in the world, primarily between the United States and China. Russia's ability to interact with the GPAI may be limited for political reasons, but, from a practical point of view, cooperation would help the country implement its national AI strategy. The Global Partnership on Artificial Intelligence (GPAI) was officially launched on June 15, 2020, at the initiative of the G7 countries alongside Australia, India, Mexico, New Zealand, South Korea, Singapore, Slovenia and the European Union. According to the Joint Statement from the Founding Members, the GPAI is an "international and multistakeholder initiative to guide the responsible development and use of AI, grounded in human rights, inclusion, diversity, innovation, and economic growth."

A former Google engineer has been sentenced to 18 months in prison after pleading guilty to stealing trade secrets before joining Uber's effort to build robotic vehicles for its ride-hailing service. The sentence handed down Tuesday by U.S. District Judge William Alsup came more than four months after former Google engineer Anthony Levandowski reached a plea agreement with the federal prosecutors who brought a criminal case against him last August. Levandowski, who helped steer Google's self-driving car project before landing at Uber, was also ordered to pay more than $850,000. Alsup had taken the unusual step of recommending the Justice Department open a criminal investigation into Levandowski while presiding over a high-profile civil trial between Uber and Waymo, a spinoff from a self-driving car project that Google began in 2007 after hiring Levandowski to be part of its team. Levandowski eventually became disillusioned with Google and left the company in early 2016 to start his own self-driving truck company, called Otto, which Uber eventually bought for $680 million. He wound up pleading guilty to one count, culminating in Tuesday's sentencing.

TikTok has become the latest platform to ban deepfakes. Under its new policy, the app says it "prohibits synthetic or manipulated content that misleads users by distorting the truth of events in a way that could cause harm." "Our intent is to protect users from things like shallow or deep fakes, so while this kind of content was broadly covered by our guidelines already, this update makes the policy clearer for our users." TikTok's General Manager Vanessa Pappas wrote in a statement. The deepfake ban is part of a broader set of policy changes meant to fight misinformation and election meddling.

The U.S. Department of Education's Institute of Education Sciences has awarded the National Center for Research on Education Access and Choice (REACH) at Tulane University a $100,000 contract to collect data from approximately 150,000 school websites across the country to see how the nation's education system is responding to the coronavirus pandemic. The project, which will track traditional public schools, charter schools and private schools, aims to quickly answer questions that are critical for understanding how students are learning when school buildings are closed. Key questions include: how many schools are providing any kind of instructional support; which are delivering online instruction; what resources are they offering to students and how do students stay in contact with teachers? "This data will also help answer important questions about equity in the school system, showing how responses differ according to characteristics like spending levels, student demographics, internet access, and if there are differences based on whether it is a private, charter or traditional public school," said REACH National Director Douglas N. Harris, Schlieder Foundation Chair in Public Education and chair of economics at Tulane University School of Liberal Arts. REACH will work in cooperation with Nicholas Mattei, assistant professor of computer science at Tulane University School of Science and Engineering, to create a computer program that will collect data from every school and district website in the country.