Cyber-Physical Mechatronic Systems such as autonomous vehicles, and intelligent transportation rely heavily on machine learning techniques for ever-increasing levels of autonomy.
In the example of autonomous vehicles, machine learning can be employed for perception, sensor fusion, resilience, prediction, and control tasks. However powerful such machine learning techniques have become, they also expose a new attack surface, which may lead to vulnerability to adversarial attacks and potentially harmful consequences in security- and safety-critical scenarios.
This PhD topic investigate machine learning challenges faced by autonomous Cyber-Physical Mechatronic Systems with the aim of formulating defence strategies.
This PhD topic is mainly composed of two parts:
• Investigating adversarial attacks. • Countermeasures The main deep learning techniques of interest to autonomous Cyber-Physical Mechatronic Systems include convolutional neural networks for detection, recurrent neural networks for time series predilection or forecasting, and deep reinforcement learning for control.
The technical innovations of the PhD topic include:
• Comprehensive survey in the direction of generating adversarial attack on deep neural network with focus on Cyber-Physical Mechatronic Systems. • Combine unsupervised learning models such as clustering (soft and hard) with an ensemble of supervised learning models to build a hybrid models that is able to able to adapt to the changes (non-stationary, anomalies) in the incoming data. • Propose a new optimization technique to provide better convergence of the machine learning algorithms. • Implement and evaluate the proposed attack and defence approaches on real-world prototypes of autonomous cyber physical systems for autonomous vehicles. • Develop an open source models, algorithms, software.