Abstract
Biological brains possess an unparalleled ability to generalize adaptive behavioral responses from only a few examples. How neural processes enable this capacity to extrapolate is a fundamental open question. A prominent but underexplored hypothesis suggests that generalization is facilitated by a low-dimensional organization of collective neural activity. Here we tested this hypothesis in the framework of flexible timing tasks where dynamics play a key role. Examining trained recurrent neural networks we found that confining the dynamics to a low-dimensional subspace allowed tonic inputs to parametrically control the overall input-output transform and enabled smooth extrapolation to inputs well beyond the training range. Reverse-engineering and theoretical analyses demonstrated that this parametric control of extrapolation relies on a mechanism where tonic inputs modulate the dynamics along non-linear manifolds in activity space while preserving their geometry. Comparisons with neural data from behaving monkeys confirmed the geometric and dynamical signatures of this mechanism.
Competing Interest Statement
The authors have declared no competing interest.