Scientific lectures: Equivalent learning emerges from distinct population dynamics across brain networks

Learning requires identifying population activity patterns associated with desired outcomes and acquiring the ability to volitionally recruit them. Whether such processes rely on shared or circuit-specific population dynamics across the brain remains unresolved. A fundamental challenge is that different brain regions typically underpin distinct behaviors, making it difficult to disentangle circuit-specific learning mechanisms from region-specific behavioral demands.
Here, we overcome this limitation using a brain-computer interface (BCI) to impose an identical, population-defined learning problem across brain networks. By making reward contingent directly on neuronal population activity, we apply the same associative learning paradigm under identical conditions to two regions with prominent recurrent connectivity but fundamentally distinct dynamical regimes: primary motor cortex (M1) and hippocampal area CA3. Mice acquired robust volitional control in both regions, demonstrating that distinct circuits can support equivalent goal-directed learning. In both M1 and CA3, learning drove progressive sparsification of population activity and greater population-level representation of the reward state. The underlying population dynamics, however, diverged. M1 showed sustained pre-reward excitation and population trajectories that flowed continuously through the reward state, with manifold geometry separating behavioral states. CA3 exhibited a learning-dependent temporal reversal, shifting from pre-reward excitation to post-reward inhibition, with population trajectories converging onto the reward state before rapidly resetting to baseline, and distinct behavioral states collapsing onto a shared manifold subspace. Recurrent network models with distinct connectivity architectures and dynamical constraints, matched to each region, recapitulated these population-level dynamics, linking intrinsic circuit structure to the geometry of learned activity.
These findings demonstrate that equivalent learning outcomes can arise from fundamentally different population-level implementations, reflecting intrinsic circuit constraints rather than conserved population-level strategies. This principled degeneracy indicates that learning emerges not from a single canonical solution, but from multiple circuit-specific implementations shaped by local dynamics.

Hosted by Jaime DE JUAN SANZ

If you would like to meet the speaker, please contact us.