Cortico-cerebellar modularity as an architectural inductive bias for efficient temporal learning

Alexandra Voce, Emmanouil Giannakakis, Claudia Clopath · 2026

Read Cortico-cerebellar modularity as an architectural inductive bias for efficient temporal learning by Alexandra Voce, Emmanouil Giannakakis, Claudia Clopath (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-10356.txt.

Description

Alexandra Voce, Emmanouil Giannakakis, and Claudia Clopath borrow a brain design to make artificial learning more efficient. The cerebellum and cerebral cortex form tight circuits thought to support flexible time processing, yet how that link shapes cortical learning is unclear. Their 2026 paper adds a cerebellar-style feedforward module to a recurrent neural network, which processes sequences over time.

The resulting cortico-cerebellar net learned faster and hit higher peak scores than size-matched fully recurrent baselines across varied regimes. Key test: freezing the recurrent core after brief training and handing later learning to the cerebellar module kept the efficiency edge. That suggests the cerebellar module drives much of the gain, with the cortical net largely acting as a fixed reservoir. Mixed modular design, the authors conclude, can serve as a strong structural bias for neural systems.

Catalog record

CREATORAlexandra Voce, Emmanouil Giannakakis, Claudia Clopath
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE52 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.10356

Browse the full library · About the Tower of Babel