Nandi Schoots
Nandi Schoots

About Me

I’m searching for research questions that cut to the core of human understanding and hope to one day find natural laws that make highly specific predictions and apply in many contexts. In the mean-time I’m working on a variety of projects that piqued my interest.

I am an FLI postdoctoral researcher in Alessandro Abate’s OXCAV group at the University of Oxford. Previously, I did my PhD in the STAI group under supervision of Peter McBurney (King’s College London) and Murray Shanahan (Imperial College London, DeepMind).

I was one of the main organizers of the first AI Safety Camp.

Interests

  • Science of Deep Learning and (Implicit) Inductive Biases in Neural Networks
  • Machine Unlearning and Activation Steering
  • Societal Impacts of AI and AI Safe
Recent Publications
(2025). Relating Piecewise Linear Kolmogorov Arnold Networks to ReLU Networks . AISTATS.
(2024). The Propensity for Density in Feed-forward Models . ECAI.
(2024). Dissecting Language Models: Machine Unlearning via Selective Pruning . SoLaR @NeurIPS.
(2024). Hidden in Plain Text: Emergence & Mitigation of Steganographic Collusion in LLMs .
(2023). Extending Activation Steering to Broad Skills and Multiple Behaviours .
(2023). Improving Activation Steering in Language Models with Mean-Centring . Responsible Language Models @AAAI.
(2023). Any Deep ReLU Network is Shallow . ECAI.
(2023). Finding Sparse Initialisations using Neuroevolutionary Ticket Search (NeTS). In ALIFE.