Deep Neural Networks (DNNs) achieve the state-of-the-art performance in several tasks in computer vision, natural language processing, image reconstruction, and many other areas. However, their black-box nature makes them inscrutable. In particular, they often learn spurious correlations between data, especially when the latter is not sufficiently varied. This problem, which is aggravated under distribution shifts (i.e., when the training and testing data differ significantly), can be traced back to the fact that all its knowledge has to be distilled from the training data. While prior knowledge, e.g., physical models, can be encoded in the training data via augmentation, embedding such information into the network's architecture has the potential to guarantee better generalization, robustness, and smaller data requirements.
In this project, which will be done in close collaboration with SeeByte Ltd, Edinburgh, we will look into generative models applied to underwater imagery. The goal is to improve the performance of current DNN-based algorithms for detection by exploiting the knowledge of the sonar acquisition process and, possibly, other physical models like motion and geometry. By designing physics-informed DNNs, we hope to obtain algorithms that not only are more reliable than conventional DNN ones, but also require smaller training datasets.
The student will be integrated in a dynamic team, supervised by Dr. Joao Mota, at Heriot-Watt University, and will interact regularly with SeeByte Ltd., a company that produces state-of-the-art software for managing unmanned and remote robotics assets in the maritime domain. The student will also be integrated into the National Robotarium, the UK's centre for Robotics and Artificial Intelligence.
All applicants must have or expect to have a 1st class degree, or equivalent, by January 2024. A good background in mathematics and programming is expected. Selection will be based on academic excellence and research potential, and all short-listed applicants will be interviewed (in person or via Teams).
Closing Date: The latest start date is January 2024. The successful candidate will commence studies as soon as possible.
Salary: As this project is co-funded by industry, there will be a top-up (around £4k/year) to the standard salary provided by EPSRC (currently around £17k/year).
How to Apply
When applying through the Heriot-Watt on-line system please ensure you provide the following information:
(a) in Study Option
You will need to select Edinburgh and Postgraduate Research. Programme presents you with a drop-down menu. Choose Electrical PhD for study option.
(b) in Research Project Information
You will be provided with a free text box for details of your research project. Enter the title of the project for which you are applying, Embedding physical models into deep neural networks for sonar detection, and also enter the supervisor’s name, Dr. Joao Mota.
This information will greatly assist us in tracking your application.
For questions about the application process, please contact email@example.com. For questions about the project, please contact Dr. Joao Mota.