A1474
Title: Tuning diagonal scale matrices for HMC
Authors: Jimmy Huy Tran - University of Stavanger (Norway) [presenting]
Tore Selland Kleppe - University of Stavanger (Norway)
Abstract: Three approaches for adaptively tuning diagonal scale matrices for HMC are discussed and compared. The common practice of scaling according to estimated marginal standard deviations is taken as a benchmark. Scaling according to the mean log-target gradient (ISG) and a scaling method targeting the frequency of when the underlying Hamiltonian dynamics crosses the respective medians should be uniform across dimensions, which are taken as alternatives. Numerical studies suggest that the ISG method leads, in many cases, to more efficient sampling than the benchmark, particularly in cases with strong correlations or non-linear dependencies. The ISG method is also easy to implement, computationally cheap, and relatively simple to include in automatically tuned codes as an alternative to benchmark practice.
