Cong Yu¹, Xin Wang¹, Cai-Xia Chen¹, Shi-Ming Tang¹, Zhong-Lin Lu²; ¹Peking University, ²New York University

The external noise paradigm (Barlow, 1957) has been widely applied in psychophysics (Pelli, 1981) and neurophysiology (Anderson-et-al, 2000) to investigate properties of human observers and neuronal tuning, as well as changes associated with attention and perceptual learning (Lu&Dosher, 2013). In this study, we employed this paradigm with two-photon calcium imaging and a transformer model to study orientation decoding from neuronal populations in macaque V1. Specifically, we simultaneously recorded the responses of more than 1,000 neurons within each FOV to a Gabor stimulus (12 orientations, contrasts = 0.03–0.50) embedded in white external noise (RMS: 0.0–0.29) in two awake, fixating macaques. We found that external noise suppressed the population orientation tuning functions, reducing the amplitude and widening the bandwidth with increasing external noise contrast. At low Gabor contrasts and high external noise, the bandwidth became unmeasurable. A gain control model provided a good fit for the observed external noise suppression effects. To decode the population responses from the recorded neurons, we applied a transformer model to reconstruct the trial-by-trial Gabor images from the neuronal response vectors. Orientation information was then extracted from the reconstructed images to evaluate orientation decoding accuracy. In a wide range of stimulus conditions, where the Gabor contrasts were not too low and the external noise levels were not too high, orientation decoding accuracy was unaffected by the contrast of external noise, despite the suppression of the population orientation tuning function. This stability was achieved by recruiting more hub neurons and establishing more effective neuronal connections. However, decoding accuracy was severely compromised under the highest external noise conditions (RMS: 0.16 & 0.29). The constant decoding accuracy across many experimental conditions, achieved through the recruitment of hub neurons and more effective connections, offers important new insights into how the visual system extracts information from noisy neuronal responses

Acknowledgements: STI2030-Major Projects grant (2022ZD0204600)