Structure of a recurrently connected artificial network receiving noisy vestibular and visual heading inputs. Comparable vestibular and visual input magnitude and signal-to-noise ratio (b, left panel), RNN output behavioral performance (b, mid and right panels), choice (c), and modality (d) patterns. The red arrow indicates that we artificially delayed the visual input (solid versus dashed Gaussian profile in the left panel of b) to reproduce the visual-choice delay in (c). The solid profiles are real inputs. The vertical dashed line in (c, d) represents stimulus offset. Colors represent sensory modalities in (b, c). Meaning of curve color and style in (d) is the same as Fig. 2e, just as shown. Higher visual input magnitude with comparable signal-to-noise ratio (e, left panel), RNN output behavioral performance (e, mid and right panels), choice (f), and modality (g) patterns.
