Computation close to criticality

We investigate neural information processing in the context of critical phenomena, because criticality has been shown to maximize information processing capacities in models. We analyze how the closeness to criticality is related to information processing in models and real biological networks. To this end, we apply wide range of technics:

  • statistical physics approach to understand which transitions in critical-like networks can be indeed considered critical
  • information-theoretical tools to evaluate data from in vitro recordings of dissociated neuronal cultures and in vivo recordings of behaving rats.
  • simulations and analysis of recurrent networks

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
  • Zierenberg, J., Wilting, J., Priesemann, V., & Levina, A. (2020). Tailored ensembles of neural networks optimize sensitivity to stimulus statistics. Physical Review Research, 2(1), 13115. https://doi.org/10.1103/physrevresearch.2.013115
  • Zierenberg, J., Wilting, J., Priesemann, V., & Levina, A. (2020). Description of spreading dynamics by microscopic network models and macroscopic branching processes can differ due to coalescence. Physical Review E, 101(2), 22301. https://doi.org/10.1103/PhysRevE.101.022301
  • Das, A., & Levina, A. (2019). Critical Neuronal Models with Relaxed Timescale Separation. Physical Review X, 9(2), 1–30. https://doi.org/10.1103/PhysRevX.9.021062