In this article, we study the problem of air-to-ground ultra-reliable and low-latency communication (URLLC) for a moving ground user. This is done by controlling multiple unmanned aerial vehicles (UAVs) in real time while avoiding inter-UAV collisions. To this end, we propose a novel multi-agent deep reinforcement learning (MADRL) framework, coined a graph attention exchange network (GAXNet). In GAXNet, each UAV constructs an attention graph locally measuring the level of attention to its neighboring UAVs, while exchanging the attention weights with other UAVs so as to reduce the attention mismatch between them. Simulation results corroborates that GAXNet achieves up to 4.5x higher rewards during training. At execution, without incurring inter-UAV collisions, GAXNet achieves 6.5x lower latency with the target 0.0000001 error rate, compared to a state-of-the-art baseline framework.

Logical reasoning, which is closely related to human cognition, is of vital importance in human's understanding of texts. Recent years have witnessed increasing attentions on machine's logical reasoning abilities. However, previous studies commonly apply ad-hoc methods to model pre-defined relation patterns, such as linking named entities, which only considers global knowledge components that are related to commonsense, without local perception of complete facts or events. Such methodology is obviously insufficient to deal with complicated logical structures. Therefore, we argue that the natural logic units would be the group of backbone constituents of the sentence such as the subject-verb-object formed "facts", covering both global and local knowledge pieces that are necessary as the basis for logical reasoning. Beyond building the ad-hoc graphs, we propose a more general and convenient fact-driven approach to construct a supergraph on top of our newly defined fact units, and enhance the supergraph with further explicit guidance of local question and option interactions. Experiments on two challenging logical reasoning benchmark datasets, ReClor and LogiQA, show that our proposed model, \textsc{Focal Reasoner}, outperforms the baseline models dramatically. It can also be smoothly applied to other downstream tasks such as MuTual, a dialogue reasoning dataset, achieving competitive results.

Machine learning applications have become ubiquitous. Their applications from machine embedded control in production over process optimization in diverse areas (e.g., traffic, finance, sciences) to direct user interactions like advertising and recommendations. This has led to an increased effort of making machine learning trustworthy. Explainable and fair AI have already matured. They address knowledgeable users and application engineers. However, there are users that want to deploy a learned model in a similar way as their washing machine. These stakeholders do not want to spend time understanding the model. Instead, they want to rely on guaranteed properties. What are the relevant properties? How can they be expressed to stakeholders without presupposing machine learning knowledge? How can they be guaranteed for a certain implementation of a model? These questions move far beyond the current state-of-the-art and we want to address them here. We propose a unified framework that certifies learning methods via care labels. They are easy to understand and draw inspiration from well-known certificates like textile labels or property cards of electronic devices. Our framework considers both, the machine learning theory and a given implementation. We test the implementation's compliance with theoretical properties and bounds. In this paper, we illustrate care labels by a prototype implementation of a certification suite for a selection of probabilistic graphical models.

Reinforcement learning (RL) is an effective technique for training decision-making agents through interactions with their environment. The advent of deep learning has been associated with highly notable successes with sequential decision making problems - such as defeating some of the highest-ranked human players at Go. In digital advertising, real-time bidding (RTB) is a common method of allocating advertising inventory through real-time auctions. Bidding strategies need to incorporate logic for dynamically adjusting parameters in order to deliver pre-assigned campaign goals. Here we discuss techniques toward using RL to train bidding agents. As a campaign metric we particularly focused on viewability: the percentage of inventory which goes on to be viewed by an end user. This paper is presented as a survey of techniques and experiments which we developed through the course of this research. We discuss expanding our training data to include edge cases by training on simulated interactions. We discuss the experimental results comparing the performance of several promising RL algorithms, and an approach to hyperparameter optimization of an actor/critic training pipeline through Bayesian optimization. Finally, we present live-traffic tests of some of our RL agents against a rule-based feedback-control approach, demonstrating the potential for this method as well as areas for further improvement. This paper therefore presents an arrangement of our findings in this quickly developing field, and ways that it can be applied to an RTB use case.

Context-aware machine translation models are designed to leverage contextual information, but often fail to do so. As a result, they inaccurately disambiguate pronouns and polysemous words that require context for resolution. In this paper, we ask several questions: What contexts do human translators use to resolve ambiguous words? Are models paying large amounts of attention to the same context? What if we explicitly train them to do so? To answer these questions, we introduce SCAT (Supporting Context for Ambiguous Translations), a new English-French dataset comprising supporting context words for 14K translations that professional translators found useful for pronoun disambiguation. Using SCAT, we perform an in-depth analysis of the context used to disambiguate, examining positional and lexical characteristics of the supporting words. Furthermore, we measure the degree of alignment between the model's attention scores and the supporting context from SCAT, and apply a guided attention strategy to encourage agreement between the two.

Detailed parcel-level crop type mapping for the whole European Union (EU) is necessary for the evaluation of agricultural policies. The Copernicus program, and Sentinel-1 (S1) in particular, offers the opportunity to monitor agricultural land at a continental scale and in a timely manner. However, so far the potential of S1 has not been explored at such a scale. Capitalizing on the unique LUCAS 2018 Copernicus in-situ survey, we present the first continental crop type map at 10-m spatial resolution for the EU based on S1A and S1B Synthetic Aperture Radar observations for the year 2018. Random forest classification algorithms are tuned to detect 19 different crop types. We assess the accuracy of this EU crop map with three approaches. First, the accuracy is assessed with independent LUCAS core in-situ observations over the continent. Second, an accuracy assessment is done specifically for main crop types from farmers declarations from 6 EU member countries or regions totaling >3M parcels and 8.21 Mha. Finally, the crop areas derived by classification are compared to the subnational (NUTS 2) area statistics reported by Eurostat. The overall accuracy for the map is reported as 80.3% when grouping main crop classes and 76% when considering all 19 crop type classes separately. Highest accuracies are obtained for rape and turnip rape with user and produced accuracies higher than 96%. The correlation between the remotely sensed estimated and Eurostat reported crop area ranges from 0.93 (potatoes) to 0.99 (rape and turnip rape). Finally, we discuss how the framework presented here can underpin the operational delivery of in-season high-resolution based crop mapping.

Various work has suggested that the memorability of an image is consistent across people, and thus can be treated as an intrinsic property of an image. Using computer vision models, we can make specific predictions about what people will remember or forget. While older work has used now-outdated deep learning architectures to predict image memorability, innovations in the field have given us new techniques to apply to this problem. Here, we propose and evaluate five alternative deep learning models which exploit developments in the field from the last five years, largely the introduction of residual neural networks, which are intended to allow the model to use semantic information in the memorability estimation process. These new models were tested against the prior state of the art with a combined dataset built to optimize both within-category and across-category predictions. Our findings suggest that the key prior memorability network had overstated its generalizability and was overfit on its training set. Our new models outperform this prior model, leading us to conclude that Residual Networks outperform simpler convolutional neural networks in memorability regression. We make our new state-of-the-art model readily available to the research community, allowing memory researchers to make predictions about memorability on a wider range of images.

Google's entry into health diagnostics has alarmed health experts who fear a new artificial intelligence tool to identify skin conditions could lead to overdiagnosis, or rare and complex skin conditions being missed. At a technology conference in the US on Tuesday, Google revealed there are almost 10bn Google searches related to skin, nail and hair issues every year. In response, Google has developed an artificial intelligence "dermatology assist tool" for people with concerns about their skin. Users of the app can use their phone to take three images of their skin, hair or nails from different angles. The app will then ask users questions about their skin type, how long they have had the issue, and for other symptoms that help narrow down the possibilities.

The Australian government is strengthening the country's digital economy through new investments in artificial intelligence (AI), cyber security and digital government services, among other areas. The investments, aimed at bolstering Australia's competitiveness in the global technology sector, are part of the government's 2021-2022 budget, which was unveiled last week. The centrepiece of the budget is arguably the A$1.2bn Digital Economy Strategy, a set of policies and actions the government is taking to grow Australia's future as a leading digital economy by 2030. But to put that expenditure into perspective, it is less than half of the A$2.6bn earmarked for a single 6km road project in Adelaide – even as it is welcomed by some quarters of the technology industry.

We introduce Dynaboard, an evaluation-as-a-service framework for hosting benchmarks and conducting holistic model comparison, integrated with the Dynabench platform. Our platform evaluates NLP models directly instead of relying on self-reported metrics or predictions on a single dataset. Under this paradigm, models are submitted to be evaluated in the cloud, circumventing the issues of reproducibility, accessibility, and backwards compatibility that often hinder benchmarking in NLP. This allows users to interact with uploaded models in real time to assess their quality, and permits the collection of additional metrics such as memory use, throughput, and robustness, which -- despite their importance to practitioners -- have traditionally been absent from leaderboards. On each task, models are ranked according to the Dynascore, a novel utility-based aggregation of these statistics, which users can customize to better reflect their preferences, placing more/less weight on a particular axis of evaluation or dataset. As state-of-the-art NLP models push the limits of traditional benchmarks, Dynaboard offers a standardized solution for a more diverse and comprehensive evaluation of model quality.