The history of how severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) spread around the planet has been far from clear. Several narratives have been propagated by social media and, in some cases, national policies were forged in response. Now that many thousands of virus sequences are available, two studies analyzed some of the key early events in the spread of SARS-CoV-2. Bedford et al. found that the virus arrived in Washington state in late January or early February. The viral genome from the first case detected had mutations similar to those found in Chinese samples and rapidly spread and dominated subsequent undetected community transmission. The other viruses detected had origins in Europe. Worobey et al. found that early introductions into Germany and the west coast of the United States were extinguished by vigorous public health efforts, but these successes were largely unrecognized. Unfortunately, several major travel events occurred in February, including repatriations from China, with lax public health follow-up. Serial, independent introductions triggered the major outbreaks in the United States and Europe that still hold us in the grip of control measures. Science , this issue p. [571][1], p. [564][2] Accurate understanding of the global spread of emerging viruses is critical for public health responses and for anticipating and preventing future outbreaks. Here we elucidate when, where, and how the earliest sustained severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) transmission networks became established in Europe and North America. Our results suggest that rapid early interventions successfully prevented early introductions of the virus from taking hold in Germany and the United States. Other, later introductions of the virus from China to both Italy and Washington state, United States, founded the earliest sustained European and North America transmission networks. Our analyses demonstrate the effectiveness of public health measures in preventing onward transmission and show that intensive testing and contact tracing could have prevented SARS-CoV-2 outbreaks from becoming established in these regions. [1]: /lookup/doi/10.1126/science.abc0523 [2]: /lookup/doi/10.1126/science.abc8169

Deep learning pioneer Yoshua Bengio has provocative ideas about the future of AI. For the first part of this article series, see here. It has only been 8 years since the modern era of deep learning began at the 2012 ImageNet competition. Progress in the field since then has been breathtaking and relentless. If anything, this breakneck pace is only accelerating. Five years from now, the field of AI will look very different than it does today.

For the first part of this article series, see here. It has only been 8 years since the modern era of deep learning began at the 2012 ImageNet competition. Progress in the field since then has been breathtaking and relentless. If anything, this breakneck pace is only accelerating. Five years from now, the field of AI will look very different than it does today. Methods that are currently considered cutting-edge will have become outdated; methods that today are nascent or on the fringes will be mainstream.

Huawei's TECH4ALL initiative aims to ensure nobody is left behind in the digital world by encouraging digital inclusion programmes and empowering technology adoption globally. The project is similar to some of the work happening within academia across Europe, where research projects are focused on harnessing technology for societal good. Professor van Ginneken, Professor of Medical Image Analysis at Radboud University Medical Centre in The Netherlands, is introducing digitized healthcare solutions to developing countries and believes that in ten years' time, all hospital pathology departments will be digitized. When did your work in medical imaging begin? I studied physics and completed a PhD in medical image analysis in 1996, developing computer programs that analyse chest x-rays using artificial intelligence (AI). At the end of the 1990s we wanted to put digital chest x-ray units with AI software in countries where there was a lot of tuberculosis, because it accommodates faster, more widespread screening, without the need to develop images on film.

Performance indicators that contributed to success at the group stage and play-off stages of the 2019 Rugby World Cup were analysed using publicly available data obtained from the official tournament website using both a non-parametric statistical technique, Wilcoxon's signed rank test, and a decision rules technique from machine learning called RIPPER. Our statistical results found that ball carry effectiveness (percentage of ball carries that penetrated the opposition gain-line) and total metres gained (kick metres plus carry metres) were found to contribute to success at both stages of the tournament and that indicators that contributed to success during the group stages (dominating possession, making more ball carries, making more passes, winning more rucks, and making less tackles) did not contribute to success at the play-off stage. Our results using RIPPER found that low ball carries and a low lineout success percentage jointly contributed to losing at the group stage, while winning a low number of rucks and carrying over the gain-line a sufficient number of times contributed to winning at the play-off stage of the tournament. The results emphasise the need for teams to adapt their playing strategies from the group stage to the play-off stage at tournament in order to be successful.

In this work we introduce a 3-valued logic with modalities, with the aim of having a clear and precise representation of epistemic states, thus the formulas of this logic will be our epistemic states. Indeed, these formulas are identified with ranking functions of 3 values, a generalization of total preorders of three levels. In this framework we analyze some types of changes of these epistemic structures and give syntactical characterizations of them in the introduced logic. In particular, we introduce and study carefully a new operator called Cautious Improvement operator. We also characterize all operators that are definable in this framework.

Learning representations of sets of nodes in a graph is crucial for applications ranging from node-role discovery to link prediction and molecule classification. Graph Neural Networks (GNNs) have achieved great success in graph representation learning. However, expressive power of GNNs is limited by the 1-Weisfeiler-Lehman (WL) test and thus GNNs generate identical representations for graph substructures that may in fact be very different. More powerful GNNs, proposed recently by mimicking higher-order-WL tests, only focus on representing entire graphs and they are computationally inefficient as they cannot utilize sparsity of the underlying graph. Here we propose and mathematically analyze a general class of structure-related features, termed Distance Encoding (DE). DE assists GNNs in representing any set of nodes, while providing strictly more expressive power than the 1-WL test. DE captures the distance between the node set whose representation is to be learned and each node in the graph. To capture the distance DE can apply various graph-distance measures such as shortest path distance or generalized PageRank scores. We propose two ways for GNNs to use DEs (1) as extra node features, and (2) as controllers of message aggregation in GNNs. Both approaches can utilize the sparse structure of the underlying graph, which leads to computational efficiency and scalability. We also prove that DE can distinguish node sets embedded in almost all regular graphs where traditional GNNs always fail. We evaluate DE on three tasks over six real networks: structural role prediction, link prediction, and triangle prediction. Results show that our models outperform GNNs without DE by up-to 15\% in accuracy and AUROC. Furthermore, our models also significantly outperform other state-of-the-art methods especially designed for the above tasks.

We investigate partial identification of causal relative and attributable risk---the ratio of two counterfactual proportions and the difference between them---in case-control and case-population studies. The odds ratio is shown to be a sharp upper bound on causal relative risk under the monotone treatment response and monotone treatment selection assumptions, without resorting to strong ignorability, nor to the rare-disease assumption. Sharp bounds on causal attributable risk are also obtained under the same assumptions. Paying special attention to the (conditional) odds ratio, we propose a semiparametrically efficient estimator of the aggregated (log) odds ratio. Further, we develop easy-to-implement causal inference procedures for relative and attributable risk. Finally, we showcase our methodology by applying it to two unique datasets in the literature. We find that attending private school may have little effect on entering a very selective university in Pakistan and that dropping out of school could substantially increase relative and attributable risk of joining a criminal gang in Brazil.

Question answering is an effective method for obtaining information from knowledge bases (KB). In this paper, we propose the Neural-Symbolic Complex Question Answering (NS-CQA) model, a data-efficient reinforcement learning framework for complex question answering by using only a modest number of training samples. Our framework consists of a neural generator and a symbolic executor that, respectively, transforms a natural-language question into a sequence of primitive actions, and executes them over the knowledge base to compute the answer. We carefully formulate a set of primitive symbolic actions that allows us to not only simplify our neural network design but also accelerate model convergence. To reduce search space, we employ the copy and masking mechanisms in our encoder-decoder architecture to drastically reduce the decoder output vocabulary and improve model generalizability. We equip our model with a memory buffer that stores high-reward promising programs. Besides, we propose an adaptive reward function. By comparing the generated trial with the trials stored in the memory buffer, we derive the curriculum-guided reward bonus, i.e., the proximity and the novelty. To mitigate the sparse reward problem, we combine the adaptive reward and the reward bonus, reshaping the sparse reward into dense feedback. Also, we encourage the model to generate new trials to avoid imitating the spurious trials while making the model remember the past high-reward trials to improve data efficiency. Our NS-CQA model is evaluated on two datasets: CQA, a recent large-scale complex question answering dataset, and WebQuestionsSP, a multi-hop question answering dataset. On both datasets, our model outperforms the state-of-the-art models. Notably, on CQA, NS-CQA performs well on questions with higher complexity, while only using approximately 1% of the total training samples.

Edge intelligence is an emerging network architecture that integrates sensing, communication, computing components, and supports various machine learning applications, where a fundamental communication question is: how to allocate the limited wireless resources (such as time, energy) to the simultaneous model training of heterogeneous learning tasks? Existing methods ignore two important facts: 1) different models have heterogeneous demands on training data; 2) there is a mismatch between the simulated environment and the real-world environment. As a result, they could lead to low learning performance in practice. This paper proposes the learning centric wireless resource allocation (LCWRA) scheme that maximizes the worst learning performance of multiple classification tasks. Analysis shows that the optimal transmission time has an inverse power relationship with respect to the classification error. Finally, both simulation and experimental results are provided to verify the performance of the proposed LCWRA scheme and its robustness in real implementation.