This work was a super fun collaboration with Mitra, advised by Guillaume Hennequin, Yashar Ahmadian, and Sonja Hofer! Thanks a lot to all of them, and to everyone else who gave us feedback! [7/7]

This mechanism, which we call dynamic consensus building, leads to a gradual reconciliation of computations across areas, and could provide a general foundation for understanding multi-area computations in the cortex. [6/7]

We developed a theoretical framework explaining how inter-area connections give rise to approximate attractor dynamics. The key insight was that reciprocal excitatory connections slow decay for congruent stimulus patterns, while accelerating it for inconsistent ones! [5/7]

Surprisingly, approximate line attractor dynamics consistently emerged in the models we fit. These dynamics allowed to sustain cortical activity even though inputs were transient, and were mediated by long-range excitatory connections. [4/7]

We fit latent input-driven dynamical system models to sets of neural recordings + optogenetic perturbations. These models were constrained by known circuit motifs, such as excitatory-only long-range projections. This ensured they remained biologically interpretable and testable. [3/7]

The cortex is highly modular, with distinct areas processing information locally. Yet, it operates as a coordinated whole. How are these modules integrated? To address this, we studied information processing between two visual cortical areas. [2/7]