Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
DY: Fachverband Dynamik und Statistische Physik
DY 34: Poster: Machine Learning, Data Science, and Reservoir Computing
DY 34.3: Poster
Mittwoch, 20. März 2024, 15:00–18:00, Poster C
Exploring neural criticality through the structure of input-induced attractors in random neural networks under external perturbations — •Hiromichi Suetani1,2 and Ulrich Parlitz3,4 — 1Faculty of Science and Technology, Oita Univerisyty, Oita, Japan — 2International Research Center for Neurointelligence, The University of Tokyo, Tokyo, Japan — 3Max Planck Research Group Biomedical Physics, Max Planck Institute for Dynamics and Self-Organization, Göttingen, Germany — 4Institute for the Dynamics of Complex Systems, University of Göttingen, Göttingen, Germany.
In recent years, a focus has turned to the neural criticality hypothesis, suggesting that the neural system optimizes information processing by maintaining activity near a critical point between order and disorder. Reservoir computing (RC) provides a theory for the neural criticality. For example, the hyperparameter region with the maximum Lyapunov exponent (LE) near zero, termed the ``edge of chaos," claims the optimality of the performance of RC. Yet, reservoirs are non-autonomous dynamical systems with external perturbations. The maximal LE is defined for autonomous systems and if applied to non-autonomous systems, it is the conditional LE where its positivity and negativity doesn't generally indicate the existence of chaos.
This study explores input-induced attractors in random neural networks under external inputs. Examining them through generalized synchronization and embeddings, we aim at developing a new theoretical foundation for neural criticality by elucidating the relationship with performances of RC such as information processing capacity.
Keywords: neural criticality; random neural networks; generalized synchronization; embeddings; reservoir computing