Posts
Some of my written posts!
Natural adversaries and out-of-distribution detection
A brief look at a more ‘natural’ approach to adversarial attacks and whether they can be considered to be ‘within distribution’ of the original model.
Making a start with feature visualisation! This is a really cool method when trying to interpret the information a network is learning.
Discusses the more advanced aspects of feature visualisation improving on the previous part!
Looks at a particular weakness of neural networks when we give them out-of-distribution data to classify.
What my PhD is about!
Feature Visualisation Part 3 (an update)
An update on feature visualisation using learned kernels! I’ve simplified some code and thrown out some unnecessary loss functions.
Feature Visualisation for Transformers (a first attempt)
A first attempt at applying the feature visualisation techniques to a vision transformer (ViT) which didn’t turn out too well!
Exploring the use of Explainable AI (XAI) as a way of understanding how models work and avoiding mistakes when training them on real data!
Hacking Pokemon for R(ea)L fun
Hacking the Pokémon Crystal Game Boy Color game to try to train an RL agent. This is part 1, discussing setting up the game to make it suitable for AI training.
Graph Embedding for Malware Detection - A failed paper
A look at a failed paper on Graph Embedding for Malware Detection. This was work I did early on in my PhD, but needed to be abandoned due to complexity and time constraints.