Groups
Hamiltonian Monte Carlo (HMC) uses gradients of the log-density to propose long-distance moves that still land in high-probability regions.
Natural gradient scales the ordinary gradient by the inverse Fisher information matrix to account for the geometry of probability distributions.