Flexible versus canonical model for contrast dependence of surround modulation with attention. Surround contrast is fixed to the highest contrast and center contrast is varied. Left: data (adapted from Sundberg et al., ); middle: canonical model; right: flexible normalization model. (a) Surround modulation index (SMI) for attending and ignoring surround, as a function of center contrast (*x*-axis) and attention location (3 curves; corresponding to surround, distant, and center attention conditions). (b) SMI for attention to the surround location minus SMI for attention to distant location. In the model panels, dark blue curves show optimal fits to Sundberg et al. () data (see , *Model parameters and fits*). For the flexible model, we also hand-changed the prior to 0.15 less and more than optimal fit (cyan), and in the canonical we also hand-changed additive constant to 2 to dominate the fit (cyan). (c) Probability of assignment. The flexible model takes a weighted sum of (divide by surround condition) and (divide only by center condition), with the weighting given by the posterior weights. In the canonical model the posterior weights are set to 1 such that there is always divisive normalization by the surround.

## PubMed Commons