Jared M. Salisbury

This study shows retinal population codes adjust their predictive timescale to input temporal structure using the information bottleneck framework.

The paper develops a dynamic scale-mixture model to analyze motion in natural scenes, finding that velocity distributions are heavy-tailed and can be modeled with a Gaussian process multiplied by a dynamic scaling factor.

The paper analyzes retinal population codes in natural scenes to show that stimulus-invariant population structure contributes to scene identity encoding.