Understanding neural circuit principles for representation learning through joint-embedding predictive architectures
Ashena Gorgan Mohammadi, Manu Srinath Halvagal, Friedemann Zenke
This paper introduces a neural circuit theory for representation learning that focuses on predicting in representation space rather than input space, using a recurrent joint-embedding predictive architecture to learn object identities and dynamics from sensory inputs.
bioRxiv · Nov 25, 2025 · PreprintNeuroscience