Figure 9:

Schematic of stimulus space, data, and adapative stimulus selection method. (A) The stimulus on each trial was a drifting Gabor patch, aligned with the cell’s preferred orientation, whose color varied from trial-to-trial. The axes of this space are the inputs to long (L), medium (M), and short (S) wavelength-sensitive cones. The center of this space represents a neutral gray stimulus with no spectral modulation. (B) Example data for one cell. The stimulus on each trial was uniquely defined by three numbers (cone contrasts for L, M, and S in the upper-diagonal half-cube) and produced a spike count responsemeasured over 667 ms. (C) Schematic of the adaptive stimulus selection procedure. On each trial, we presented stimulus xt and recorded spike response r_t. Next, we updated the posterior over the tuning curve g using the full data set collected so far, ${\mathcal{D}}_t$. Finally, we computed the posterior variance of the tuning curve, $\text{var}(\lambda ({\mathbf{x}}^{\ast })\mid {\mathcal{D}}_t)$, for a grid of points x* spanning the stimulus space, and selected the point with maximal variance as the stimulus for the next trial.