Huge congratulations to STAI CDT’s very first graduate, Dr Patrick Henriksen, on successfully defending his thesis, ‘Robust Neural Networks: Verification, Training, and Repair’.
We spoke to Patrick about his PhD thesis, being part of STAI CDT and what his future plans are.
Well done on passing your PhD viva! What is your thesis about?
My PhD thesis is titled “Robust Neural Networks: Verification, Training, and Repair”. Neural networks are a class of machine learning algorithms which are particularly well suited for handling complex high-dimensional data in domains such as audio, image, or video processing.
Some applications include voice assistants such as Siri and Alexa, large language models such as ChatGPT, and video processing in self-driving vehicles. The term “robust” here refers to the networks’ ability to perform well under disturbances of their inputs; such disturbances could arise from environmental noise, shifting lighting conditions, or deteriorating sensors.
My thesis proposes methods for formal verification of neural networks, i.e., mathematically proving the networks’ robustness to disturbances. It also proposes methods for training neural networks to become more robust to disturbances. Finally, the thesis proposes a method for repairing undesired behaviour of neural networks post-training.
How have you found being part of the STAI CDT?
I have really enjoyed my time at the STAI CDT. I particularly appreciated the social aspect of doing a PhD with so many like-minded students and staff. The CDT also provided me with the opportunity to attend a host of exciting seminars, activities and retreats that complement my research.
What are your future plans now that you’ve passed your PhD viva?
I’ve now started working full-time in Safe Intelligence, an Imperial College London spinout, where I’ll continue to do research and implement methods revolving around the robustness of machine learning algorithms.
We look forward to seeing your career progress outside the STAI CDT and are excited that you are now STAI CDT’s first alumni. Congratulations again Dr Patrick Henriksen.