Drone technology continues to advance, as more research and development is targeted toward traffic control systems for the small, flying devices. The Nevada Institute for Autonomous Systems (NIAS) was recently awarded the bulk of a roughly $1.8 million earmark by the Federal Aviation Administration (FAA) to study and test virtual unmanned traffic management technology, known as UTM. The effort is a partnership among NIAS, an FAA-designated drone test site; Switch, maker of data-center technology; and ANRA Technologies, which produces drones. Switch and ANRA will lead demonstrations and testing of unmanned flight systems, while NIAS will explore some of the system and requirements to operate drone fleets safely. Advances in drone traffic control could not be more timely, say makers of the devices, as companies explore using drones for any number of on-demand deliveries -- from groceries to chicken wings.

Concerned that the majority of air quality issues affected those in low-income countries, the team developed a solution called AirQo, an initiative that combines human ingenuity, AI models, and boxes packed with air monitoring technology to predict pollution patterns in Kampala. Air sensors on buildings and moto-taxis collect swathes of pollution data, and cloud-based AI software swiftly analyses it to make air quality forecasts. These predictions are passed to government agencies, who can work to improve air quality and reduce the risk of exposure within local communities. The research team hopes that one day this technology will reduce pollution on streets across the continent so that this generation and the next will know what it means to breathe fresh air.

AI can be used to enhance the accuracy and efficiency of decision-making and to improve lives through new apps and services. It can be used to solve some of the thorny policy problems of climate change, infrastructure and healthcare. It is no surprise that governments are therefore looking at ways to build AI expertise and understanding, both within the public sector but also within the wider community. To unleash the potential of AI safely, however, issues such as accuracy, human control, transparency, bias and privacy need to be addressed. So governments should be role-modelling the ethical use of AI, and educating their people on AI and how to be ready for the opportunities and challenges. One way countries could do this would be through setting up a body that is a visible focus for AI: a centre of excellence.

As 12,000 lightning strikes pummeled the Bay Area this month, igniting hundreds of fires, fire spotters sprang into action. Their arsenal of tools includes thermal imagery collected by space satellites; real-time feeds from hundreds of mountaintop cameras; a far-flung array of weather stations monitoring temperature, humidity and winds; and artificial intelligence to munch and crunch the vast data troves to pinpoint hot spots. For decades, wildfires in remote regions were spotted by people in lookout towers who scanned the horizon with binoculars for smoke -- a tough and tedious job. They reported potential danger by telephone, carrier pigeon or Morse code signals with a mirror. Now, fire spotting has gone high tech.

Open-set recognition and adversarial defense study two key aspects of deep learning that are vital for real-world deployment. The objective of open-set recognition is to identify samples from open-set classes during testing, while adversarial defense aims to defend the network against images with imperceptible adversarial perturbations. In this paper, we show that open-set recognition systems are vulnerable to adversarial attacks. Furthermore, we show that adversarial defense mechanisms trained on known classes do not generalize well to open-set samples. Motivated by this observation, we emphasize the need of an Open-Set Adversarial Defense (OSAD) mechanism. This paper proposes an Open-Set Defense Network (OSDN) as a solution to the OSAD problem. The proposed network uses an encoder with feature-denoising layers coupled with a classifier to learn a noise-free latent feature representation. Two techniques are employed to obtain an informative latent feature space with the objective of improving open-set performance. First, a decoder is used to ensure that clean images can be reconstructed from the obtained latent features. Then, self-supervision is used to ensure that the latent features are informative enough to carry out an auxiliary task. We introduce a testing protocol to evaluate OSAD performance and show the effectiveness of the proposed method in multiple object classification datasets. The implementation code of the proposed method is available at: https://github.com/rshaojimmy/ECCV2020-OSAD.

Quantum computers operate in the high-dimensional tensor product spaces and are known to outperform classical computers on many problems. They are poised to accelerate machine learning tasks in the future. In this work, we operate in the quantum machine learning (QML) regime where a QML model is trained using a quantum-classical hybrid algorithm and inferencing is performed using a quantum algorithm. We leverage the traditional two-step machine learning workflow, where features are extracted from the data in the first step and a discriminator acting on the extracted features is used to classify the data in the second step. Assuming that the binary features have been extracted from the data, we propose a quantum discriminator for binary classification. The quantum discriminator takes as input the binary features of a data point and a prediction qubit in the zero state, and outputs the correct class of the data point. The quantum discriminator is defined by a parameterized unitary matrix $U_\Theta$ containing $\mathcal{O}(N)$ parameters, where $N$ is the number of data points in the training data set. Furthermore, we show that the quantum discriminator can be trained in $\mathcal{O}(N \log N)$ time using $\mathcal{O}(N \log N)$ classical bits and $\mathcal{O}(\log N)$ qubits. We also show that inferencing for the quantum discriminator can be done in $\mathcal{O}(N)$ time using $\mathcal{O}(\log N)$ qubits. Finally, we use the quantum discriminator to classify the XOR problem on the IBM Q universal quantum computer with $100\%$ accuracy.

The primary objective of supervised learning is to develop models that generalize robustly to unseen data. Benchmark test sets provide a proxy for out-of-sample performance, but can outlive their usefulness in some cases. For example, evaluating on benchmarks alone may steer us towards models that adaptively overfit [Reu03; RFR08; Dwo 15] to the finite test set and do not generalize. Alternatively, we might select for models that are sensitive to insignificant aspects of the dataset creation process and thus do not generalize robustly (e.g., models that are sensitive to the exact set of humans who annotated the test set). To diagnose these issues, recent work has generated new, previously "unseen" testbeds for standard datasets through a process known as dataset replication. Though not yet widespread in machine learning, dataset replication is a natural analogue to experimental replication studies in the natural sciences (cf.

Graph Neural Networks (GNNs) are powerful and flexible neural networks that use the naturally sparse connectivity information of the data. GNNs represent this connectivity as sparse matrices, which have lower arithmetic intensity and thus higher communication costs compared to dense matrices, making GNNs harder to scale to high concurrencies than convolutional or fully-connected neural networks. We introduce a family of parallel algorithms for training GNNs and show that they can asymptotically reduce communication compared to previous parallel GNN training methods. We implement these algorithms, which are based on 1D, 1.5D, 2D, and 3D sparse-dense matrix multiplication, using torch.distributed on GPU-equipped clusters. Our algorithms optimize communication across the full GNN training pipeline. We train GNNs on over a hundred GPUs on multiple datasets, including a protein network with over a billion edges.

The COviD-19 Ontology for cases and patient information (CODO) provides a model for the collection and analysis of data about the COVID-19 pandemic. The ontology provides a standards-based open-source model that facilitates the integration of data from heterogeneous data sources. The ontology was designed by analysing disparate COVID-19 data sources such as datasets, literature, services, etc. The ontology follows the best practices for vocabularies by re-using concepts from other leading vocabularies and by using the W3C standards RDF, OWL, SWRL, and SPARQL. The ontology already has one independent user and has incorporated real-world data from the government of India.

Robots can be used to collect environmental data in regions that are difficult for humans to traverse. However, limitations remain in the size of region that a robot can directly observe per unit time. We introduce a method for selecting a limited number of observation points in a large region, from which we can predict the state of unobserved points in the region. We combine a low rank model of a target attribute with an information-maximizing path planner to predict the state of the attribute throughout a region. Our approach is agnostic to the choice of target attribute and robot monitoring platform. We evaluate our method in simulation on two real-world environment datasets, each containing observations from one to two million possible sampling locations. We compare against a random sampler and four variations of a baseline sampler from the ecology literature. Our method outperforms the baselines in terms of average Fisher information gain per samples taken and performs comparably for average reconstruction error in most trials.