Rui Ponte Costa
@somnirons.bsky.social
Computational neuroscientist bringing machine and neural learning closer together [Oxford NeuroAI] @ox.ac.uk @oxforddpag.bsky.social 🇵🇹🇪🇺🇬🇧🌳
neuralml.github.io
www.medsci.ox.ac.uk/neuroai
neuralml.github.io
www.medsci.ox.ac.uk/neuroai
Yes, but that 'target' is sensory information right? So in practice this 'target' is new sensory data that the brain will try to predict/match. Of course that in practice we model these tasks things using SL, but this is an abstraction, in reality the brain doesn't have *direct* access to targets.
January 7, 2026 at 3:46 PM
Yes, but that 'target' is sensory information right? So in practice this 'target' is new sensory data that the brain will try to predict/match. Of course that in practice we model these tasks things using SL, but this is an abstraction, in reality the brain doesn't have *direct* access to targets.
Btw there is also related work by the group of Alex Cayco-Gajic:
www.biorxiv.org/content/10.1...
www.biorxiv.org/content/10.1...
Distributed learning across fast and slow neural systems supports efficient motor adaptation
Adaptation is a fundamental aspect of motor learning. Intelligent systems must adapt to perturbations in the environment while simultaneously maintaining stable memories. Classic work has argued that ...
www.biorxiv.org
January 7, 2026 at 3:04 PM
Btw there is also related work by the group of Alex Cayco-Gajic:
www.biorxiv.org/content/10.1...
www.biorxiv.org/content/10.1...
The key difference seems to be more on the speed of learning. We show that its only when the cerebellum learns fast and the cortex slow (fixed) we can explain data. Thus cortex=stable world 'model' and the cerebellum=fast adapter. (2/2)
www.nature.com/articles/s41...
www.nature.com/articles/s41...
Cerebellar-driven cortical dynamics can enable task acquisition, switching and consolidation - Nature Communications
How the brain maintains a stable world model while swiftly adapting to environmental changes remains unclear. Here, the authors propose that the cerebellum drives cortical dynamics, enabling rapid tas...
www.nature.com
January 7, 2026 at 3:04 PM
The key difference seems to be more on the speed of learning. We show that its only when the cerebellum learns fast and the cortex slow (fixed) we can explain data. Thus cortex=stable world 'model' and the cerebellum=fast adapter. (2/2)
www.nature.com/articles/s41...
www.nature.com/articles/s41...
Thanks Juan! I do think that virtually all forms of learning in the brain are self-supervised (so unsupervised), as there never is never an explicit target in the brain. (1/2)
January 7, 2026 at 3:04 PM
Thanks Juan! I do think that virtually all forms of learning in the brain are self-supervised (so unsupervised), as there never is never an explicit target in the brain. (1/2)
Indeed now you are about to consolidate all that knowledge you started with ;)
December 12, 2025 at 7:37 AM
Indeed now you are about to consolidate all that knowledge you started with ;)
Fantastic Juan! Exciting times!
December 10, 2025 at 10:57 PM
Fantastic Juan! Exciting times!