Simple and exact method to improve the efficiency of Hamiltonian Monte Carlo and No-U-Turn-Sampler. When simulating trajectories of Hamiltonian dynamics for a proposal, don't discard the intermediate states but recycle them! #BAjournal bit.ly/advanceBA

A numerical method for solving differential equations can be considered to be a point estimator, but why not be Bayesian instead? Can Bayesian methods for the numerical solution of differential equations be developed? It all comes down to Lie symmetries, according to Wang et al.

HMC versus Metropolis-Hastings