Grad student trying to understand the history of humanity, the future of AI, and how to make both of these things work together in the present.
Oxford, UK
Joined June 2025
- Tweets24
- Following6
- Followers82
- Likes39
"Thou shalt have the power to degenerate"
Deep Learning Is Singular - Here’s What That Means
Neural networks are full of degeneracies - many different parameter settings that implement exactly the same function. Classical statistics assumes these away. Singular learning theory puts them at the centre of the story of how neural networks learn. Matthew Farrugia-Roberts @MatthewFdashR is a lecturer at the Iliad Intensive and PhD Candidate at the University of Oxford. Here he explains why degeneracy may be the key to understanding where structure in neural networks comes from, and how singular learning theory connects to interpretability, inductive biases, and controlling what your network learns.
8
1,607
Super excited to have been part of preparing and delivering teaching materials for the Singular Learning Theory day. Thanks to Iliad and my coauthors @FurmanZach and Kai Ogden for making this intensive happen.
Have you ever tried to look inside the run folders that W&B makes for every deep learning experiment?
Here's a deep dive about how I spent a few weeks freeing 118 GB of experimental archives from undocumented and corrupted binary .wandb files:
far.in.net/free-wandb
2
126
I'm thrilled to be a part of delivering the first course on AI Safety and Alignment at the University of Oxford! Next week is going to be intense and I'm looking forward to it!
🚨New AI Safety Course @aims_oxford!
I’m thrilled to launch a new called AI Safety & Alignment (AISAA) course on the foundations & frontier research of making advanced AI systems safe and aligned at @UniofOxford
what to expect 👇
robots.ox.ac.uk/~fazl/aisaa/
3
249
At least for me, the big-picture motivation behind our RLC paper is a research vision for scalable AI alignment via minimax regret autocurricula.
Learn about the paper via co-author @Karim_abdelll: 🧵👉teapaw.org/Karim_abdelll/status/1…
Learn about why I think this is important work 🧵👇
For more complex environments, we still need better UED methods.
But UED is young!
There are plenty of plausible directions for improving over the methods that have been proposed so far.
The question is, is there enough room for improvement for this to help when it counts?
1
7
163
Our paper, "Mitigating goal misgeneralization via minimax regret," will appear at RLC 2025!
Congratulations to my co-authors @Karim_abdelll , @usmananwar391, @hannaherlebach, @casdewitt, @DavidSKrueger, and @MichaelD1729 🎉
Preprint out now arxiv.org/abs/2507.03068
Thread soon!
6
205
There are many important social and ethical issues raised by today’s AI technologies. It's also true that as we project developments in AI technology into the future, we can foresee new and different ethical issues that might arise.
1
1
121
Accordingly, last year, I was invited to give a guest lecture on ethical questions raised by potential future advancements in AI for the final week of @UniMelb's COMP90087 The Ethics of Artificial Intelligence.
youtu.be/-DvCQAiX2QA
1
153