Self-organization of neuronal networks

Collective dynamics and emergence

Selected presentations and publications:

  • S Khajehabdollahi, J Prosi, E Giannakakis, G Martius, A Levina (2022) When to be critical? Performance and evolvability in different regimes of neural Ising agents, Artificial Life, 28 (4), 458-478
  • Jan Prosi*, Sina Khajehabdollahi*, Emmanouil Giannakakis, Georg Martius, Anna Levina (2021),The dynamical regime and its importance for evolvability, task performance and generalization. ALIFE proceedings 2021
  • Khajehabdollahi, S., Martius, G., & Levina, A. (2019). Assessing Aesthetics of Generated Abstract Images Using Correlation Structure. 2019 IEEE Symposium Series on Computational Intelligence, arXiv vesrion:
  • Felix Effenberger, Juergen Jost, Anna Levina (2015). Self-organization in Balanced State Networks by STDP and Homeostatic Plasticity. PLoS Computational Biology, 11(9), 1–30.
  • Levina, A., Herrmann, J. M., & Geisel, T. (2007). Dynamical synapses causing self-organized criticality in neural networks. Nature Physics, 3(12), 9.

Subproject. Population/evolutionary dynamics

Most if not all biological systems are large collectives of components that interact in complex networks and structures to give rise to emergent group behaviours. Using relatively simple neural networks subject to evolutionary algorithms we investigate the tendency for populations of neural networks to self-organize to a critical regime. Understanding the conditions in which evolution can lead to criticality and the corresponding emergent behaviour can allow us to begin to understand evolution in a more universal framework. Furthermore, in the context of The Critical Brain Hypothesis, this project aims to understand the utility and emergent behaviours associated with critical systems.

Subproject. Neural Shaders

Utilizing randomness to generate complexity or realism in art and simulations of reality has been practiced for a number of decades in the movie, game, and digital art industries. In this project, we study how randomness can give rise to complex structures and the perception of beauty. By generating ensembles of random compositional pattern producing networks (CPPNs) and analyzing their image statistics in relation to perceptual scores of beauty, we investigate how randomness can be used to elicit the perception of complex structures and aesthetics.