Evolutionary Convergence of Hierarchical Information Processing
Viola Priesemann - MPI for Dynamics and Self-Organization, GöttingenMichael Wibral - Georg-August-Universität Göttingen
We will analyze neural data from Drosophila, Zebrafish, Mouse and Macaque to show that across lineages an anatomical hierarchical organization of a neural system is exploited in convergent ways for processing information. Anatomical hierarchical organization here is defined minimally by an ordering of separable neural groups into a processing chain that connects sensory inputs and motor outputs over at least 3 groups. Such an anatomical hierarchical organization is found in all but the simplest nervous systems. Functionally, we expect intrinsic network timescales to increase across hierarchical levels, enabling the integration of input over increasingly longer time windows; and the hierarchy to be also reflected in specific information theoretic fingerprints, indicative of predictive processing. Our core hypotheses are:(H1) that neural hierarchies across lineages functionally exhibit hierarchically increasing intrinsic network timescales of neural processing, and(H2) that higher levels of the hierarchy with long timescales exhibit a so-called reverberating state with close-to-critical dynamics convergently across lineages - as we have already shown for macaque monkey, cat, and rat;(H3) that the hierarchy of timescales is paralleled by a similar hierarchy of memory properties - measured information-theoretically as the amount of active information storage and the time window contributing to this storage, and allowing for the integration of information over increasing timescales;(H4) that hierarchies are exploited for some form of predictive processing, where the information selected for further processing at a specific level of the hierarchy is dependent on the current predictability of its inputs - showing as a specific correlation of information storage and transfer;(H5) that the above hierarchical properties change in a convergent way across lineages when the organism transits from an idling/resting state to active task performance.Formulating these hypotheses exclusively in terms of timescales and information-theoretic quantities allows us to analyze information processing in local circuits independently of the specific content that is processed. Thus, we can bridge the enormous inter-lineage differences in natural behavior, and typical experiments to demonstrate functional convergence. For example, our approach allows to compare hierarchical neural information processing in species from different lineages that focus on very different primary sensory inputs due to the respective ecological niches they inhabit. Specifically, we will analyze timescales, memory properties and predictive processing and compare across the four species. We expect that all these species show functional convergence in the sense of SPP2205 with respect to exploiting anatomical hierarchies for information integration across multiple timescales and for predictive processing.