An Analytical Framework for Multi-Area Balanced Networks
Abstract
A hallmark of neocortical architecture is recurrent connectivity both within and between local sub-networks (cortical areas). Within a cortical area, excitation-inhibition balance (balanced positive and negative connections) shapes neural activity dynamics, while reciprocal connections between areas are excitatory. How this multi-area structure shapes neural dynamics remains largely unknown. We present an analytical framework for balanced multi-area networks, revealing key features of cortical computation; we find that local connectivity within an area determines its responses to inputs received locally (extra-cortical), but not to inputs relayed from other cortical areas. Local responses to these relayed inputs are instead primarily driven by long-range inter-area connections. Moreover, we find that the asymmetry of inter-area connections (feedforward vs feedback strength) can modulate the joint dynamics across areas and implement a tradeoff between regimes that promote similarity or divergence of activity across areas.