Evolutionary Convergence of Hierarchical Information Processing

Viola Priesemann – MPI for Dynamics and Self-Organization, Universität Göttingen
Michael Wibral – Universität Göttingen

SPP 2205 calls for demonstrations of functional-level convergence between neuronal circuit operations in distinct animal lineages.One intense current debate on the neurophysiological substrates of functional-level convergence is the question how birds achieve their amazing cognitive capabilities. These capabilities are on par with those of mammals and can reach, for example in macaws, the level seen in non-human primates -- yet birds have a fundamentally different brain architecture that lacks a neocortex, which is thought to be central for mammalian cognition. This begs the question how the convergence on the level of cognitive function is supported by the avian brain, and whether a hidden convergence at the level of neuronal circuit operation can be found between mammalian cortical areas, or layers, and the modules of the avian pallium. To address questions on functional-level convergence on the level of circuit operation in the first funding phase of this SPP, we developed a framework comprised of intrinsic- and information-timescale analyses, and information-theoretic measures that allowed us to quantitatively characterize computational hierarchies across species with vastly different brain architectures. We found for example longer timescales for brain areas higher in the known mammalian cortical hierachy. For funding phase two, we will extend this framework using partial information decomposition (PID), a recent development from information theory. PID can information-theoretically characterize computation proper, rather than mere communication across information channels. Despite being a recent development PID has already proven useful in the understanding of computation in deep neural networks. Using our framework from funding phase 1, extended by PID, we will characterize cortical areas and layers in mice, as well as pallial modules in pigeons (based on data from our collaborators in Bochum) by their correlation and information timescales and their partial information decomposition - thus obtaining their fingerprints of circuit-level operation. We will then align cortical areas, or layers, and pallial modules in the space of these timescale-, information-, and PID-measures to uncover the putative alignment at the level of circuit operation between these structures. If found, such an alignment will indicate a convergence at the level of circuit operations that may underlie the functional convergence of mammals and birds with respect to their cognitive capabilities.