Thanks; good question! I would imagine that looking into coding principles that take more of the cell-type-specific computational properties or (potential) functional tasks into account should be interesting. Without excluding other possibilities.

Thanks so much to Dimos Karamanlis @dimokaramanlis.qoto.org.ap.brid.gy for spearheading this, and thanks for a great team effort to get the primate retina recordings established and carried out!

Subunit models (here with a new modeling framework, applying flickering gratings for parameter fitting) can capture much of the spatial-contrast sensitivity and response correlations. 7/

The nonlinear, spatial-contrast-sensitive mechanisms are even strong enough to induce positive activity correlations between ON and OFF parasol cells in marmosets, but not midget cells (and ON and OFF transient alpha cells, but not sustained alpha cells in mice). 6/

Correlated activity is seen particularly during fixations with high-contrast spatial structure; spatial nonlinearities of receptive fields are largely responsible for triggering the correlated activity. 5/

The correlated activity leads to redundancy in stimulus encoding, e.g., for parasol cells in marmosets or transient Off-alpha cells in mice, challenging the efficient coding hypothesis. 4/

For some ganglion cell types (e.g., OFF midgets in marmoset retina), activity of cell pairs is strongly decorrelated compared to stimulus correlations (as predicted by efficient coding); other types display hardly any decorrelation at all (e.g., ON parasols). 3/

We recorded spiking responses from populations of retinal ganglion cells in marmoset and mouse retina in response to photographs shifted according to species-specific gaze trajectories. We identified ganglion cell types, and assessed their activity correlations and coding redundancy. 2/