Fractional Brownian motion (fBM) is a stochastic process used to model anomalous diffusion, where particles spread at rates not captured by standard Brownian motion. The model is characterised by two parameters: the Hurst exponent H, which controls the roughness or memory of the trajectory, and a generalised diffusion coefficient D_H, which sets the overall scale of movement. Estimating both parameters from data is a common but nontrivial task.
In this preprint, the authors propose a maximum likelihood approach to jointly estimate H and D_H. Maximum likelihood is a standard statistical technique that selects parameter values making the observed data most probable. The abstract does not provide details on the estimator's performance or the derivation, but it frames the work as a joint inference problem rather than estimating each parameter separately.
The title also mentions neural networks and posterior mean, suggesting the full paper may compare MLE with these alternative inference strategies. However, the abstract itself only describes the maximum likelihood contribution, so the relationship between these methods is not clear from the available text.