There's a great talk by Juan Gallego on how low-dimensional #NeuralManifolds arise from biological constraints, remain invariant across states and inputs, and support cross-animal alignment. Examples span #HeadDirection rings, #gridcell tori, #MotorCortex prep vs movement, striatal timing dynamics, and C. elegans #behavior loops. Cool talk as it shows how #manifold-level structure can generalize across tasks and organisms.
馃 New preprint by Codol et al. (2025): Brain-like #NeuralDynamics for #behavioral control develop through #ReinforcementLearning. They show that only #RL, not #SupervisedLearning, yields neural activity geometries & dynamics matching monkey #MotorCortex recordings. RL-trained #RNNs operate at the edge of #chaos, reproduce adaptive reorganization under #visuomotor rotation, and require realistic limb #biomechanics to achieve brain-like control.
