Efficient Distance Approximation for Structured High-Dimensional Distributions via Learning

We design efficient distance approximation algorithms for several classes of structured high-dimensional distributions. Specifically, we show algorithms for the following problems: - Given sample access to two Bayesian networks and over known directed acyclic graphs and having nodes and bounded in-degree, approximate to within additive error using samples and time - Given sample access to two ferromagnetic Ising models and on variables with bounded width, approximate to within additive error using samples and time - Given sample access to two -dimensional Gaussians and , approximate to within additive error using samples and time - Given access to observations from two causal models and on variables that are defined over known causal graphs, approximate to within additive error using samples, where and are the interventional distributions obtained by the intervention on and respectively for a particular variable . Our results are the first efficient distance approximation algorithms for these well-studied problems. They are derived using a simple and general connection to distribution learning algorithms. The distance approximation algorithms imply new efficient algorithms for {\em tolerant} testing of closeness of the above-mentioned structured high-dimensional distributions.
View on arXiv