Designing self-adaptive network with thermodynamics

For real-world complex system constantly enduring perturbation, to achieve survival goal in changing yet unknown environments, the central problem is designing a self-adaptation strategy instead of fixed control strategies, which enables system to adjust its internal multi-scale structure according to environmental feedback. Inspired by thermodynamics, we develop a self-adaptive network utilizing only macroscopic information to achieve desired landscape through reconfiguring itself in unknown environments. By continuously estimating environment entropy, our designed self-adaptive network can adaptively realize desired landscape represented by topological measures. The adaptability of this network is achieved under several scenarios, including confinement on phase space and geographic constraint. The adaptation process is described by relative entropy corresponding to the Boltzmann H function, which decreases with time following unique power law distinguishing our self-adaptive network from memoryless systems. Moreover, we demonstrate the transformability of our self-adaptive network, as a critical mechanism of complex system resilience, allowing for transitions from one target landscape to another. Compared to data-driven methods, our self-adaptive network is understandable without careful choice of learning architecture and parameters. Our designed self-adaptive network could help to understand system intelligence through the lens of thermodynamics.