Are you interested in Machine Learning and Signal Processing methods and their applications to autonomous systems and surveillance? We have a PhD project funded by Dstl at the University of Sheffield, a World Top 100 university and a Russell Group university in the UK. This PhD stipend is significantly higher than the PhD market average so it is an excellent opportunity for professional growth and acquiring machine learning skills in a rapidly developing area and apply them to important engineering applications for sensor management and distributed systems. Modern sensor networks for defence and security can generate very large amounts of data. These networks are often comprised of sensors with different modalities, such as radar, acoustic sensors, LIDAR, combined with optical and thermal cameras. Moreover, data arrive at different rates and with varying levels of accuracy. Making sense of such multiple heterogeneous data is a challenging task that has been extensively studied, but the provision of reliable solutions for autonomous and semi-autonomous systems is a task that remains only partially solved. In autonomous surveillance systems, machine-learning algorithms play a crucial role since their outputs are integrated into various downstream tasks. The development of learning frameworks that include an awareness of their limitations and have the capacity for insightful introspection in changeable environments is an essential part of this project. These data-driven algorithms must assess their performance, predict incipient breakdown, and continually learn from large streaming datasets. In addition, these systems must operate within available energy budget in (near) real-time scenarios. This project therefore aims to develop trustworthy methods for autonomous inference, situation awareness and sensor management that are robust, interpretable and actionable in complex situations. A special emphasis will be given on Gaussian process methods that are able to learn their hyperparameters and operate under a range of changeable conditions – not only with different measurement noise uncertainties, but also environmental. Sensor networks formed by mobile and static sensors will be considered, with different types of sensors, including cloud computing. Reinforcement learning methods are also potential candidates for the considered sensor management tasks. Main project objectives: 1) to develop approaches for inference and tracking that are modular, and scalable with respect to the volume of data arising from number of targets (and so states estimated), data sampling rates, and numbers of sensors and their geographic extent; 2) to quantify the impact of uncertainties with respect to measurement noise, dynamical change in the environment, abrupt target manoeuvres among other factors; to provide reliable solutions, with a defined level of trust - these uncertainty-aware algorithms can self-assess performance and continually learn from the available datasets while operating in real-time on a limited energy budget; 3) to develop methods for sensor management and data fusion linked with inference and intent prediction, jointly applied with tracking. The project is multidisciplinary and requires knowledge from signal processing, sensor networks, probabilistic methods and machine learning. Supervised by Prof. Lyudmila Mihaylova and in support of a Dstl expert, the PhD student will be based at the Department of Automatic Control and Systems Engineering at the University of Sheffield. For additional information, please feel free to contact Prof. Mihaylova ( l.s.mihaylovasheffield.ac.uk ) with your CV and relevant qualifications / transcripts. The PhD position is available with an immediate start. - please see advert