Neural and Machine Learning Group — University of Oxford
Research Programme
01 — Neural geometry for interpretability
Methods for inspecting the internal representations that neural networks build as they learn. I study how individual units specialise, how differently specialised units come together to create a population code, and how the resulting geometry shapes behaviour — then track how all of this changes with learning. The goal is to make the hidden workings of a network legible: to tell, for example, whether a system has learned a genuine, transferable concept or just a shortcut that happens to work.
2026
02 — Interpretable control of large models
Methods for steering a large network by attaching a small, separate controller. Rather than retraining or rewiring the whole model, the controller learns to guide its dynamics toward desired behaviour while the underlying network remains fixed. This makes a fixed system flexible: it can be redirected to new tasks, switch strategies, or adapt on the fly through a controller that is far cheaper to train and study than the model itself. I pursue these ideas in two directions — as a window onto continual learning in animals and as a route toward safe, interpretable control of AI systems.
2026
03 — A digital twin of an insect brain
A working replica of a real neural circuit, built from the fruit fly neurobiology. Its wiring comes from a complete synapse-level brain map, while each cell’s electrical properties are constrained by the genes it expresses. The result is a simulated circuit one can run and dissected at scale. It also lets us apply interpretability tools from modern AI research — ablating components to find the minimal circuit that performs a function, and tracing how activity gives rise to behaviour. I hope to use it to ask how dopamine neurons reconfigure the wider circuit to allow rapid adaptation.
2026
Curriculum vitae
2023–present
Postdoctoral Research Scientist in Machine Learning, AI Behaviour and Computational Neuroscience
Neural and Machine Learning Group, University of Oxford
2022–2023
Research Assistant in Cognitive Computational Neuroscience
Cognitive Computational Neuroscience Laboratory, University of Oxford
2019–2024
DPhil in Computational Neuroscience
University of Oxford, UK
2016–2018
Research Scientist in Electrophysiology
Nencki Institute of Experimental Biology, Polish Academy of Sciences
selected Publications
Granule cells reorient cortical manifolds to separate contexts but preserve their geometry
Garcia-Garcia, M. G., Wojcik, M. J., … & Wagner, M. J.
2026
Learning shapes neural geometry in the prefrontal cortex
Wójcik, M. J., Stroud, J. P., … & Stokes, M. G.
Nature Neuroscience
2026
Organizing across disciplines to tackle shared computational challenges
Treyde, W., Kwiatkowski, A., Acherberg J., … & Wójcik, M. J.
Patterns
2026
Working memory shapes neural geometry in human EEG over learning
Wójcik, M. J., … Myers, N. E., & Hunt, L. T.
eLife · 14, RP106609
2025
Effects of noise and metabolic cost on cortical task representations
Stroud, J. P., Wojcik, M. J., … & Lengyel, M.
eLife · 13, RP94961
2025