Rapid response to natural hazards, such as floods, is essential to mitigate loss of life and the reduction of suffering. For emergency response teams, access to timely and accurate data is essential. Satellite imagery offers a rich source of information which can be analysed to help determine regions affected by a disaster. Much remote sensing flood analysis is semi-automated, with time consuming manual components requiring hours to complete. In this study, we present a fully automated approach to the rapid flood mapping currently carried out by many non-governmental, national and international organisations. We design a Convolutional Neural Network (CNN) based method which isolates the flooded pixels in freely available Copernicus Sentinel-1 Synthetic Aperture Radar (SAR) imagery, requiring no optical bands and minimal pre-processing. We test a variety of CNN architectures and train our models on flood masks generated using a combination of classical semi-automated techniques and extensive manual cleaning and visual inspection. Our methodology reduces the time required to develop a flood map by 80%, while achieving strong performance over a wide range of locations and environmental conditions. Given the open-source data and the minimal image cleaning required, this methodology can also be integrated into end-to-end pipelines for more timely and continuous flood monitoring.

In an unprecedented effort of scientific collaboration, researchers across fields are racing to support the response to COVID-19. Making a global impact with AI tools will require scalable approaches for data, model and code sharing; adapting applications to local contexts; and cooperation across borders.

Precision theoretical predictions for high multiplicity scattering rely on the evaluation of increasingly complicated scattering amplitudes which come with an extremely high CPU cost. For state-of-the-art processes this can cause technical bottlenecks in the production of fully differential distributions. In this article we explore the possibility of using neural networks to approximate multi-variable scattering amplitudes and provide efficient inputs for Monte Carlo integration. We focus on QCD corrections to e ⁺ e ⁻ → jets up to one-loop and up to five jets. We demonstrate reliable interpolation when a series of networks are trained to amplitudes that have been divided into sectors defined by their infrared singularity structure. Complete simulations for one-loop distributions show speed improvements of at least an order of magnitude over a standard approach.

Machine learning technology has the potential to dramatically optimise event generation and simulations. We continue to investigate the use of neural networks to approximate matrix elements for high-multiplicity scattering processes. We focus on the case of loop-induced diphoton production through gluon fusion, and develop a realistic simulation method that can be applied to hadron collider observables. Neural networks are trained using the one-loop amplitudes implemented in the NJet C++ library, and interfaced to the Sherpa Monte Carlo event generator, where we perform a detailed study for 2 → 3 and 2 → 4 scattering problems. We also consider how the trained networks perform when varying the kinematic cuts effecting the phase space and the reliability of the neural network simulations.

X-Ray image enhancement, along with many other medical image processing applications, requires the segmentation of images into bone, soft tissue, and open beam regions. We apply a machine learning approach to this problem, presenting an end-to-end solution which results in robust and efficient inference. Since medical institutions often do not have the resources to process and label the large quantity of X-Ray images usually needed for neural network training, we design an end-to-end solution for small datasets, while achieving state-of-the-art results. Our implementation produces an overall accuracy of 92%, F1 score of 0.92, and an AUC of 0.98, surpassing classical image processing techniques, such as clustering and entropy based methods, while improving upon the output of existing neural networks used for segmentation in non-medical contexts. The code used for this project is available online.¹

We introduce J une , an open-source framework for the detailed simulation of epidemics on the basis of social interactions in a virtual population constructed from geographically granular census data, reflecting age, sex, ethnicity and socio-economic indicators. Interactions between individuals are modelled in groups of various sizes and properties, such as households, schools and workplaces, and other social activities using social mixing matrices. J une provides a suite of flexible parametrizations that describe infectious diseases, how they are transmitted and affect contaminated individuals. In this paper, we apply J une to the specific case of modelling the spread of COVID-19 in England. We discuss the quality of initial model outputs which reproduce reported hospital admission and mortality statistics at national and regional levels as well as by age strata.