Abstract
How are simple perceptual decisions formed based on noisy neural signals that are distributed over large populations of neurons in early sensory cortical areas? To begin to address this fundamental question, we used a combination of voltage-sensitive dye imaging and electrophysiology to measure directly neural population responses in the primary visual cortex (V1) of monkeys while they performed a reaction-time visual detection task. We then evaluated the ability of different candidate decoding models to detect the target from the measured neural responses. Our analysis reveals that previously proposed models for pooling neural responses over space and time are highly inefficient given the statistics of V1 population responses. We derived the optimal Bayesian decoder of V1 responses and show that it could be implemented by simple neural circuits. Finally, we show that the optimal decoder outperforms the monkey in both speed and accuracy, indicating that inefficiencies at, or downstream to, V1 limit performance in simple detection tasks.
This work was supported by National Eye Institute grants EY016454 and EY016752 and by a Sloan Foundation Fellowship