Neurodynamics of Computation

Lab Description

The Neurodynamics of Computation group studies how brain circuits compute, and in particular how they keep learning over a lifetime without losing what they have already learned. We approach this as reverse engineering. We build detailed biophysical models of real circuits, starting from the anatomy and physiology of the olivocerebellar system, and use them as a plausibility filter: of the many principles proposed for learning and control, we ask which ones still hold once the biology is taken seriously. The group works closely with the experimental labs of the Department of Neuroscience and is part of the Neurocomputing Lab, which Mario Negrello co-leads with Christos Strydis.

Research lines

The cerebellum as a continual learner. Skilled behaviour depends on the cerebellum learning the dynamics of whatever body and task it is coupled to, and on keeping that calibration accurate as both change. Our model of the olivocerebellar loop suggests that biology handles this through its architecture, with homeostatic, bidirectional plasticity in upbound and downbound micromodules (Fernández Santoro et al., PLOS Computational Biology, 2025). With colleagues in the department we have also studied how inhibitory and excitatory inputs gate the output of the inferior olive, and how Purkinje cells shape coherence between sensory and motor cortex. A current aim is to distil a minimal, differentiable learning rule from the detailed models, and then to find out which pieces of biological detail have to be put back for the rule to keep working.

Continual learning from brains to machines. Artificial networks still tend to overwrite old knowledge when they learn something new, while cerebellar and hippocampal circuits manage to keep both. We carry learning rules from our circuit models into machine-learning settings, build benchmarks grounded in motor learning, and test the results in embodied systems. Related theoretical work with Peter König (Osnabrück) proposes the transformer as a computational analogy for the organisation of the cortical microcircuit (König & Negrello, Cognitive Neurodynamics, 2026).

Simulation tools. Biophysical models at this scale need fast simulation. Together with the Neurocomputing Lab we develop EDEN, a high-performance neural simulator built on NeuroML, as well as hardware libraries for biophysically detailed models. Mario Negrello also leads the digital-twins work package of a Convergence consortium of TU Delft, Erasmus MC and Erasmus University.

About the PI

Mario Negrello is Associate Professor and Principal Investigator in the Department of Neuroscience at Erasmus MC. After training as a mechanical engineer at the Federal University of Paraná in Brazil and several years in industry, he moved to the University of Osnabrück, where he obtained an M.Sc. and, in 2009, a PhD in Cognitive Science with the highest distinction. His doctoral work on the artificial evolution of neural controllers for autonomous robots was carried out at the Fraunhofer Institute in Sankt Augustin, with a research stay in Hod Lipson's lab at Cornell University. A postdoc in Erik De Schutter's Computational Neuroscience Unit at OIST took him into large-scale cerebellar modelling, and from there to Rotterdam. His monograph Invariants of Behavior appeared with Springer in 2012. He has co-supervised seven doctoral theses, serves as Associate Editor of Frontiers in Computational Neuroscience, and in 2023 co-founded CONNECT, the Dutch computational neuroscience and neurotechnology network, with Jorge Mejias and Fleur Zeldenrust.

Teaching is a large part of the group's life. Mario Negrello founded and runs NeuroAI: The Computing Brain, a 30 EC minor for bachelor students at Erasmus University Rotterdam, teaches computational science in the Nanobiology programme of TU Delft and Erasmus MC, and coordinates computational neuroscience in the Erasmus MC Research Master in Neuroscience.

Find out more

Our research themes, team, publications and source code are on the Neurocomputing Lab website: Research · People · Publications · Source code · Vacancies. Students and researchers from neuroscience, physics, engineering, mathematics or computer science who would like to work with us are welcome to get in touch.