Learning Multivariate Log-concave Distributions

We study the problem of estimating multivariate log-concave probability density functions. We prove the first sample complexity upper bound for learning log-concave densities on Rd, for all d ≥ 1. Prior to our work, no upper bound on the sample complexity of this learning problem was known for the case of d > 3. In more detail, we give an estimator that, for any d ≥ 1 and ǫ > 0, draws Õd ( (1/ǫ)(d+5)/2 ) samples from an unknown target log-concave density on Rd, and outputs a hypothesis that (with high probability) is ǫ-close to the target, in total variation distance. Our upper bound on the sample complexity comes close to the known lower bound of Ωd ( (1/ǫ)(d+1)/2 ) for this problem.