SAMOP 2023 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
Q: Fachverband Quantenoptik und Photonik
Q 7: Poster I
Q 7.32: Poster
Monday, March 6, 2023, 16:30–19:00, Empore Lichthof
Network self-organization: the role of the activation function — •Frederic Folz1, Kurt Mehlhorn2, and Giovanna Morigi1 — 1Theoretische Physik, Universität des Saarlandes, 66123 Saarbrücken, Germany — 2Algorithms and Complexity Group, Max-Planck-Institut für Informatik, Saarland Informatics Campus, 66123 Saarbrücken, Germany
The interplay of nonlinear dynamics and noise is at the basis of coherent phenomena, such as stochastic resonance, synchronization, and noise-induced phase transitions. In a recent work we analysed network dynamics in the presence of Gaussian noise when the activation function, governing the dynamics of the network connections, is a sigmoidal (Hill) function. In these settings we demonstrated the onset of network topologies that maximize the transport efficiency and behave as noise-induced resonances. In this work we systematically analyse this behavior for different classes of activation functions. We identify the activation function that yields the most robust network configuration at the the maximal convergence speed. We then discuss possible applications to neural networks in the quantum domain.