Meshulam, L. | Search bioRxiv

bioRxiv2026

Hierarchical processing and polarization encoding in the cephalopod visual system

Coleoid cephalopods (octopus, cuttlefish and squid, hereafter cephalopods) have evolved a range of complex visually-guided behaviours, from dexterous hunting to skin-pattern based camouflage and communication1. They have also evolved sensitivity to the polarization of light2, an adaptation thought to help detect camouflaged or semitransparent predators and prey in low visibility underwater environments3-10. How visual information is processed by the cephalopod brain to support their behaviours remains unclear. Here, studying the bigfin reef squid Sepioteuthis lessoniana, we performed calcium imaging and electrophysiological recordings from populations of neurons in the large visual center of the cephalopod brain, the optic lobe (OL). We revealed that the retina-recipient superficial OL contains a diversity of functionally distinct cell types, spatially organized into sub-layers, processing spatio-temporal features of light intensity and possessing polarization angle specificity. More complex features, e.g. direction selectivity, are seen in deeper regions of the OL cortex, which also exhibits spontaneous waves of neural activity in the absence of visual input. Neurons in the downstream OL medulla exhibit visual receptive field sizes and spontaneous activity levels which increase with brain depth, consistent with the hierarchical processing of visual information through the medullas tree-like anatomical organization. Medulla neurons exhibit sensitivity to local decreases in the degree of linear polarization (DoLP), which they integrate additively with light intensity information. Underwater imaging in the squids habitat off the coast of Okinawa, Japan, demonstrate that polarization sensitivity confers a robust short-range boost in object-background contrast over a range of objects and environmental conditions. These findings reveal convergent principles of hierarchical visual processing shared between cephalopods and vertebrates, and highlight how cephalopods utilise their distinct adaptation of polarization sensitivity to solve universal visual challenges underwater.

neuroscience↗

bioRxiv2020

Reverse-engineering Recurrent Neural Network solutions to a hierarchical inference task for mice

We study how recurrent neural networks (RNNs) solve a hierarchical inference task involving two latent variables and disparate timescales separated by 1-2 orders of magnitude. The task is of interest to the International Brain Laboratory, a global collaboration of experimental and theoretical neuroscientists studying how the mammalian brain generates behavior. We make four discoveries. First, RNNs learn behavior that is quantitatively similar to ideal Bayesian baselines. Second, RNNs perform inference by learning a two-dimensional subspace defining beliefs about the latent variables. Third, the geometry of RNN dynamics reflects an induced coupling between the two separate inference processes necessary to solve the task. Fourth, we perform model compression through a novel form of knowledge distillation on hidden representations - Representations and Dynamics Distillation (RADD)- to reduce the RNN dynamics to a low-dimensional, highly interpretable model. This technique promises a useful tool for interpretability of high dimensional nonlinear dynamical systems. Altogether, this work yields predictions to guide exploration and analysis of mouse neural data and circuity.

neuroscience↗