Neural wave interference in inhibition-stabilized networks

To gain insight into the neural events responsible for visual perception of static and dynamic optical patterns, we study how neural activation spreads in arrays of inhibition-stabilized neural networks with nearest-neighbor coupling. The activation generated in such networks by local stimuli propagates between locations, forming spatiotemporal waves that affect the dynamics of activation generated by stimuli separated spatially and temporally, and by stimuli with complex spatiotemporal structure. These interactions form characteristic interference patterns that make the network intrinsically selective for certain stimuli, such as modulations of luminance at specific spatial and temporal frequencies and specific velocities of visual motion. Due to the inherent nonlinearity of the network, its intrinsic tuning depends on stimulus intensity and contrast. The interference patterns have multiple features of “lateral” interactions between stimuli, well known in physiological and behavioral studies of visual systems. The diverse phenomena have been previously attributed to distinct neural circuits. Our results demonstrate how the canonical circuit can perform the diverse operations in a manner predicted by neural-wave interference. Author Summary. We developed a framework for analysis of biological neural networks in terms of neural wave interference. Propagation of activity in neural tissue has been previously studied with regards to waves of neural activation. We argue that such waves generated at one location should interfere with the waves generated at other locations and thus form patterns with predictable properties. Such interactions between effects of sensory stimuli are commonly found in behavioral and physiological studies of visual systems, but the interactions have not been studied in terms of neural wave interference. Using a canonical model of the inhibitory-excitatory neural circuit, we investigate interference of neural waves in one-dimensional chains and two-dimensional arrays of such circuits with nearest-neighbor coupling. We define conditions of stability in such systems with respect to corrugation perturbations and derive the control parameters that determine how such systems respond to static, short-lived, and moving stimuli. We demonstrate that the interference patterns generated in such networks endow the system with many properties of biological vision, including selectivity for spatial and temporal frequencies of intensity modulation, selectivity for velocity, “lateral” interactions between spatially and temporally separate stimuli, and predictable delays in response to static and moving stimuli. 1/35 ar X iv :1 41 0. 42 37 v3 [ qbi o. N C ] 1 S ep 2 01 6