The vertebrate retina performs prediction on incoming visual signals, which can compensate for lags in neural processing [1]. It has been hypothesized that prediction occurs at each successive layer of the visual stream, but downstream prediction of the retina is not yet understood. One challenge is that retinal responses are collective, and full recovery of predictive information must take into account correlations in the joint activity of large populations; this incurs the curse of dimensionality. Another challenge arises when the stimulus is naturalistic, since relevant features in complex scenes are typically unknown. Furthermore, estimates of available predictive information in complex scenes and the responses they elicit are difficult. In this work, we address these challenges simultaneously by searching for maximally-predictive collective variables in a large population of 93 salamander retinal ganglion cells under naturalistic stimulus. To achieve this, we apply a tractable, approximate implementation of the information bottleneck method to our neural data [2], and infer a lower-dimensional representation that is maximally informative about the future neural activity. We observe that across stimuli and intervals, all predictive information in the retinal outputs captured by a few (less than 10) linear collective variables. We further show that predictive signals are collectively encoded. At short timescales, this coding is less collective and noise correlations contribute significantly, while at later timescales predictive features are highly collective and stimulus-induced correlations dominate. Our analysis demonstrates the feasibility of finding biologically relevant coarse-graining schemes in high-dimensional data using variational inference and basic machine learning tools.
[1] S. E. Palmer, O. Marre, M. J. Berry, and W. Bialek, “Predictive information in a sensory population,” Proceedings of the National Academy of Sciences, vol. 112, no. 22, pp. 6908–6913, Jun. 2015
[2] D. E. Gökmen, Z. Ringel, S. D. Huber, and M. Koch-Janusz, “Symmetries and phase diagrams with real-space mutual information neural estimation,” Phys. Rev. E, vol. 104, no. 6, p. 064106, Dec. 2021
Hosts: Michael Abbott and Jose Betancourt Valencia
