An Investigation of Embedded Machine Learning for Space Applications: Circuit Reliability and Model Security
Intelligent spacecrafts that use Machine Learning to ensure autonomous exploration would be very useful for exploring new parts of the Solar System and reducing mission costs. The explicit need for adaptability, reasoning, and generalization from past experience provides a prime opportunity for the field of ML to be deployed in such systems. However, Space missions operate in an extremely challenging and aggressive environment. In fact, soft errors such as Single Event Upset (SEU) and Single Event Transient (SET) which are typically caused by high energy cosmic particles striking electronic devices are significantly present in the space environment. On the other hand, ML models have been successfully used in wireless communication, which is a crucial requirement for space missions. However, ML models have been shown vulnerable to adversarial attacks, which consist of low magnitude maliciously crafted noise that forces the ML models to misclassify. Due to these risks, ensuring the reliability and trustworthiness of ML is a critical challenge towards the deployment of autonomous space devices.
In this project, we will investigate the inherent fault-tolerance of embedded ML devices in the context of aggressive space environment.
Our objectives are as follows:
1) Characterize the existing ML models’ inherent robustness to memory and computational errors
2) Propose comprehensive reliability enhancement techniques based on the previous exploratory analysis
3) Investigate the threat of adversarial attacks on ML models for wireless space communication.
4) Propose circuit-aware defence against adversarial noise towards trustworthy autonomous spacecrafts
This project is funded through the CSIT Doctoral Training Programme. The scholarships are fully funded, with an additional stipend for 42 months and an industrial top-up worth £27k per student. You will also benefit from the opportunity to be considered for a 1-to-3 months placement in industry with one of our partners, as well as enhanced training in leadership, professional skills and much more. Since this is a single scholarship being advertised across a number of projects, we expect a large number of competitive applicants.
For full details on the funding/training package available, and candidate eligibility criteria, please visit https://www.qub.ac.uk/ecit/CSIT/Cyber-AIHub/
The minimum academic requirement for admission is normally an Upper Second Class Honours degree from a UK or ROI Higher Education provider in a relevant discipline, or an equivalent qualification acceptable to the University.