A photo of me!

Welcome to my web page. I’m Daniel Fentham a researcher in the School of Computer Science at the University of Birmingham. My research focuses on security and interpretability of AI models. I’m currently working on a RISE funded project titled “Combating Fake Media with Edit-Tolerant Authentication” with Mark Ryan, Xiao Yang, Shize Deng and David Oswald.

Thesis title: Bayesian Geometric Deep Learning for Cyber Security. The particulars of my thesis proposal can be found here (this is a bit old now and we’ve changed direction since this was written!).

Official Thesis title: Diversity as a Defence Against Neural Network Adversarial Examples | This has been submitted and I’ve passed my viva! I guess I’m a doctor now…

My supervisors are Professor Mark Ryan and Professor Dave Parker!

This website aims to be a collection of things I’ve learned over the years and need to remember! This includes posts which contribute directly or indirectly to my PhD, along with some other fun things if I find the time. I’ve been pretty bad at posting in the past since I haven’t been comfortable with a piece if I haven’t thoroughly checked it through multiple times to ensure everything I’m saying is correct. This adds a lot of mental overhead which means that I just don’t post. This year (2024) I’ve decided to take a different approach where I’ll be adding blog posts in any state, with the idea that these will act as living documents which change and evolve as I learn new things, correct mistakes and develop different opinions. So, just like with any LLM, I hope that these posts are interesting, but don’t blindly believe them without doing some research of your own!

I’m now on ORCiD! You can see my official publications here.

Past Projects


AndroCFG

Control flow graph extraction from the Smali intermediate language which is commonly used during Android app dissasembly.

github


RMAggNet

An ensemble-based adversarial defence that promotes diversity in computer vision models by providing each with a different classification task.

github - paper


Posts

subscribe via RSS