Université de Strasbourg

Fellows seminar - Weird architectures: dynamic and structural complexity as a resource for brain function

January 17, 2023
From 15:00 until 16:30
Salle de Table Ronde, MISHA, Strasbourg

Image: Pixabay

By Demian Battaglia, 2020 Fellow

We like to think that systems implementing a function must have an architecture specially designed to perform the considered function. Somewhat as a corollary to this first belief, we are also confident that the function implemented by a system could be inferred by the observation of its architecture. This is true when referring to man-made architectures (cf. Louis Sullivan's "Form follows function" statement) and also for common ways to conceptualize the operation of biological systems and, specifically, neural systems. However, life and neural information processing are the outcome of intricate interactions between a large number of parts, whose resulting behaviour is ultimately due to collective coordination and could radically differ from the individual behaviour of one of the parts in isolation. Such collective auto-organization tends to give rise to complex and haphazard patterns and causality chains, in which one could hardly identify any sign of engineering-like rational design.

In this talk, I will present some examples of "neural architectures" at different scales: cerebellar connectivity maps, and oscillatory organization in the hippocampus and cortex, supposed to be tailored for precise functions, such as, respectively, motor coordination or the flexible routing of information across different brain regions. When averaging over many different connectivity maps or over a large number of oscillatory events, some characteristic features seem to emerge and theories have been proposed linking each of these specific features to different putative functions. However, when considered one by one, these maps and these oscillatory events are extremely diverse and irregular, frequently displaying properties very far from the ideal ones summarized by averages. The diversity is such that one may be tempted to say that the averages... do not really exist in the reality of individual circuits or in the here and now of actual neural dynamics related to ongoing behaviour.

What this extreme diversity could mean? That architectures are an idealization and that brain circuits and dynamics are disordered and largely random? Or, on the contrary, that architectures associated to function still exist, but that we have to accept them being "weird", unlike our Cartesian rationality would like them to be? And, if this is the case, which language could we use to describe "weirdness", to understand its origin and, perhaps, its computational advantages?

In collaboration with Romain Goutagny, Vincent Douchamps, Samy Castro (LNCA, UniStra) and Philippe Isope (INCI, UniStra)

France 2030