New pre-print: https://t.co/G5h3saofpr. The information bottleneck offers useful but imperfect insights into deep learning. @mircomusolesi, Steve Hailes, and I introduce a Generalized IB that fills the gaps left by the classic approach.

If you're at @aistat, join me at Spot 151 in Session 2 (Sun, May 4) to discuss my work with @mircomusolesi & Stephen Hailes: Partial Information Decomposition for Interpretability. Paper:

Updated version of pre-print at:

Pleased to announce that Partial Information Decomposition for Data Interpretability and Feature Selection will appear in AiStats'25. Please give the latest version a read at: https://t.co/Ndi3DrtEhn. @mircomusolesi

I'm happy to announce a new pre-print! In this paper, @mircomusolesi, Steve Hailes, and I introduce MIPP: a pruning technique that preserves the mutual information between layers. Explore the work here: https://t.co/VXbkR9vRyP.

Delighted to share my new pre-print with @mircomusolesi & Stephen Hailes. We've developed a new framework for selecting features and interpreting their interactions, with particularly exciting use cases in genetics and neuroscience. Explore our findings: https://t.co/Ndi3DrtEhn

🚨New paper 🚨 What does it mean to create rational machines? We survey the role of rationality and irrationality within artificial intelligence, and discuss research challenges and open questions in this area. Co-author: @mircomusolesi https://t.co/R7mvKAAMJB

Many of the current problems with AI systems arise due to the lack of social & ethical intelligence. How could - and how should we - develop moral reasoning in artificial agents? Our answer is in our latest preprint: https://t.co/olzWwwAKty - w/ @mircomusolesi & Steve Hailes 1/7

My new pre-print with Stephen Hailes and @mircomusolesi on information-theoretic state variable selection for reinforcement learning is now out: https://t.co/7Q28a34uNQ