Spatially-embedded recurrent neural networks: what can they teach us?

24 April 2024

Colloquium by Dr. Danyal Akarka, MRC Cognition and Brain Sciences Unit, University of Cambridge

Abstract:
Brain networks exist within the confines of metabolic and spatial limits while simultaneously implementing its required information processing. However, most computational work to date does not model how these competing structural and functional constraints are traded-off in real time. In his talk, Dr Danyal Akarca will introduce an approach that aims to captures these effects: called spatially embedded recurrent neural networks.

These networks learn basic task-related inferences while existing within a three-dimensional Euclidean space, where the communication of constituent neurons is constrained by a sparse connectome. These networks are found to converge on structural and functional features that are also commonly found in primate cerebral cortices. Specifically, they converge on solving inferences using modular small-world networks, in which functionally similar units spatially configure themselves to utilise an energetically efficient mixed-selective code. As these features emerge in unison, these networks can reveal how structural and functional brain motifs can be intertwined and attributed to basic optimisation processes. Danyal will close by discussing plausible directions of where this research direction may lead to.

Reference: Achterberg, J. & Akarca, D. et al. Spatially embedded recurrent neural networks reveal widespread links between structural and functional neuroscience findings. Nat Mach Intell 5, 1369–1381 (2023). https://doi.org/10.1038/s42256-023-00748-9.

More information can be found in the attached file and on the website https://www.danakarca.com/.

Date and time: 24 April 2024, 3:00 p.m., W36, Seminar Room, UKE and via Zoom