Ensemble slice sampling

Slice Sampling has emerged as a powerful Markov Chain Monte Carlo algorithm that adapts to the characteristics of target distribution with minimal hand-tuning. However, Sampling's performance is highly sensitive user-specified initial length scale hyperparameter and method generally struggles poorly scaled or strongly correlated distributions. This paper introduces Ensemble (ESS), new class algorithms bypasses such difficulties by adaptively tuning utilising an ensemble parallel walkers in order efficiently handle strong correlations between parameters. These affine-invariant are trivial construct, require no hand-tuning, can easily be implemented computing environments. Empirical tests show improve efficiency more than magnitude compared conventional MCMC methods on broad range In cases multimodal distributions, sample even high dimensions. We argue parallel, black-box gradient-free nature renders it ideal for use scientific fields physics, astrophysics cosmology which dominated wide variety computationally expensive non-differentiable models.