Back

Critical period plasticity enables credit assignment

Meier, R. J.; Muller, S. Z.; Wang, B.; Mizerska, K.; Jain, S.; Fothergill, T.; Mercer, J.; Klapheke, C.; DiSano, J.; Milicic, N.; Wang, R.; Narayan, S.; Li, J.; Weiss, K.; Alvarez-Salvado, E.; Ahrens, M. B.; Hibi, M.; Huisken, J.; Eliceiri, K. W.; Ehrlich, D. E.

2026-06-29 neuroscience
10.64898/2026.06.24.734069 bioRxiv
Show abstract

Synaptic plasticity is often guided by instructive inputs to neural circuits, but learning only succeeds when these instructions reach neurons that mediate relevant outputs. This creates the credit assignment problem: how does a neuron receive instructions suited to its own behavioral function? Here we report a developmental mechanism that enables credit assignment by using instructive inputs to organize the downstream architecture through which learning is expressed. In the olivocerebellar learning system, the inferior olive provides instructive inputs that guide plasticity within the cerebellum. During circuit assembly in zebrafish, we find these same inputs regulate long-range cerebellar projections during a highly plastic, two-day critical period. Developmental experience caused specific maturation of projections to targets that were coactivated with olivary inputs. Mathematical theory and computational modeling show how the resulting architecture constrains later learning, such that a single input can sculpt downstream connectivity and then leverage that circuit to assign credit. Simulated learning became paradoxically more robust if we reduced plasticity after a developmental critical period, protecting key architecture from corruption during learning. Thus, instructive inputs can first build the circuits they later teach, coordinating development and learning to enable effective credit assignment.

Matching journals

The top 4 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.