Back to the Team Page
Team
Postdocs
Pau Vilimelis Aceituno
Postdoc
Bio
I am a theoretician at the interface between Neuroscience, Machine Learning and Neuromorphic Computing. Using Information Theory, Signal Processing, Random Matrices and Graph Theory, I look for theories and principles that explain how neural networks learn and compute in brains and machines. I also have a general interest (and occasional projects) in other fields such as ecology or economics. Previously I have worked on designing satellites, software for error-prone hardware, data mining for airline IT, and time-series processing for microbiomes.
Publications
The underlying structures of self-attention: symmetry, directionality, and emergent dynamics in Transformer training
2025, ICML
Temporal horizons in forecasting: a performance-learnability trade-off
2025, Transactions in Machine Learning Research
Directed and acyclic synaptic connectivity in the human layer 2-3 cortical microcircuit
2024, Science
Theoretical principles explain the structure of the insect head direction circuit
2024, eLife
Bio-inspired, task-free continual learning through activity regularization.
2023, Biological Cybernetics
Learning Cortical Hierarchies with mporal Hebbian Updates.
2023, Frontiers. Comp. Neurosc
Mini-batching ecological data to improve ecosystem models with machine learning
2022, Methods in Ecology and Evolution
Credit Assignment in Neural Networks through Deep Feedback Control
2021, NeurIPS