Efficient Coding in the Economics of Human Brain Connectomics

Abstract

In systems neuroscience, most models posit that brain regions communicate information under constraints of efficiency. Yet, metabolic and information transfer efficiency across structural networks are not understood. In a large cohort of youth, we find metabolic costs associated with structural path strengths supporting information diffusion. Metabolism is balanced with the coupling of structures supporting diffusion and network modularity. To understand efficient network communication, we develop a theory specifying minimum rates of message diffusion that brain regions should transmit for an expected fidelity, and we test five predictions from the theory. We introduce compression efficiency, which quantifies differing trade-offs between lossy compression and communication fidelity in structural networks. Compression efficiency evolves with development, heightens when metabolic gradients guide diffusion, constrains network complexity, explains how rich-club hubs integrate information, and correlates with cortical areal scaling, myelination, and speed-accuracy trade-offs. Our findings elucidate how network structures and metabolic resources support efficient neural communication.

The growth and form of knowledge networks by kinesthetic curiosity

“…Last, we predict that macro-scale brain network structure and function moderate individual differences in learning related to the compressibility of knowledge networks [82,23,24,25]. We hypothesize the involvement of network hubs in the frontoparietal circuit, which are thought to support executive function, learning, and information compression, as well as the default-mode network associated with mind wandering and hierarchical abstraction [63,97,98,99,85]. A set of brain regions across both networks are associated with the reinforcement learning of implicit and explicit representations of experience [100,95].…”

“…Third, if kinesthetic curiosity is linked with hierarchical abstraction, then the integrated compressibility that corresponds to finer and coarser abstractions of the same knowledge network will explain how the cost of information motivates discounts according to distance and time. Together, the frameworks of kinesthetic curiosity and predictive processing can be bridged by examining whether knowledge networks are built to be increasingly compressible [62,63,69,24,25,23,85].…”

Edgework: Viewing curiosity as fundamentally relational

“…The fact that looser and more fluid, as opposed to tighter and more crystalline, knowledge networks may be easier to reconfigure, could explain why children are able to better learn and flexibly use information from abstract schema than adults [70]. Evidence from neuroscience further suggests that the patterns of inter-regional connections in the human brain may increasingly prioritize less compressed transmission (less abstract) as development occurs [87], potentially altering the sorts of inter-conceptual connections made and carried with a mind into later life.…”

“…Ongoing efforts seek to understand where in the brain we represent each node and edge in the network, as well as where (and how) we encode the network's mesoscale and global architecture. How are these representations and encodings dynamically created, modified, and (sometimes) forgotten [95,87]? Following Dewey, we might ask how "such a network of interconnections" is employed to "offer a point of advantage from which to get at the problem presented in a new experience" [9].…”

Neurophysiological evidence for cognitive map formation during sequence learning

“…Studies investigating representations of spatial environments have pointed out the usefulness of low dimensional representations for learning to navigate [18,22]. Evidence for dimensionality reduction of neural signals has been observed in neural structures at 3 distinct scales: single neurons, anatomical regions, and the whole brain [75][76][77][78]. Broadly, dimensionality reduction of neural signals is thought to enable the brain to easily extract important, often changing information and facilitating the development of a sparse, efficient neural code for items in the environment [77,79].…”