Bidirectional Mamba state-space model for anomalous diffusion

Abstract Characterizing anomalous diffusion (AnDi) is crucial in order to understand the evolution of complex stochastic systems, from molecular interactions to cellular dynamics. Here, we evaluate the performances regarding such a task of Bi-Mamba, a novel state-space deep-learning architecture articulated around a bidirectional scanning mechanism. Our implementation is tested on the AnDi-2 challenge datasets. Designed for regression tasks, the Bi-Mamba architecture infers efficiently the effective diffusion coefficient and anomalous exponent from single, short trajectories. As such, our results indicate the potential practical use of the Bi-Mamba architecture towards the characterization of AnDi.