Posterior convergence rates of Dirichlet mixtures at smooth densities

By Subhashis Ghosal, A. van der Vaart. 2007

235 citations recorded by the source corpus.

We study the rates of convergence of the posterior distribution for Bayesian density estimation with Dirichlet mixtures of normal distributions as the prior. The true density is assumed to be twice continuously differentiable. The bandwidth is given a sequence of priors which is obtained by scaling a single prior by an appropriate order. In order to handle this problem, we derive a new general rate theorem by considering a countable covering of the parameter space whose prior probabilities satisfy a summability condition together with certain individual bounds on the Hellinger metric entropy. We apply this new general theorem on posterior convergence rates by computing bounds for Hellinger (bracketing) entropy numbers for the involved class of densities, the error in the approximation of a smooth density by normal mixtures and the concentration rate of the prior. The best obtainable rate of convergence of the posterior turns out to be equivalent to the well-known frequentist rate for integrated mean squared error n -2/5 up to a logarithmic factor.

Subfields: Artificial Intelligence, Statistics and Probability

View Paper: https://www.semanticscholar.org/paper/dbaea7ddacc2aa81e01254b7073c536bc0b5ed0f

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Statistics and Probability

Citations in the corpus

Bayesian fractional posteriors (2016, background): “(48) From Lemma 10 of [22], for some constants C11, C12 > 0 and all sufficiently large n, Π  K ∑”
On Bayesian supremum norm contraction rates (2013, background): “nt on the second moment of log(f/f0) to further get a rate ζn for klog(f/f0)k∞. Since (15) is assumed, one can restrict to the event {f:h(f,f0) ≤ εn} ⊂ {f:kf−f0k2 .εn}. We have (see, e.g., Lemma 8 in [14]), klog(f/f0)k 2 2 .h(f,f0)(1+logkf/f0k∞). The last term is bounded by a constant times ε2 n(1+l”
Rates of contraction for posterior distributions in Lr-metrics, 1 ≤ r ≤ ∞ (2011, background): “This led to further progress recently; we refer to [11, 12, 32, 34] and the references therein.”
Rates of contraction of posterior distributions based on Gaussian process priors (2008, background): “The second and third inequalities now follow from the first by Lemma 8 of [10].”
Noname manuscript No. (will be inserted by the editor) A semiparametric Bernstein- von Mises theorem for Gaussian process priors (unknown): “An application of Lemma 8 in [11] then gives that the second moment W (P0, P) verifies W (P0, P) h(2)(P0, P)nε2 n nε4 n .”
Thomas Bayes’ walk on manifolds (2012, background): “By Lemma 8 in [13], for any densities p, q”
Bayesian inverse problems with non-conjugate priors (2012, background): “Our results therefore mirror those in Theorem 1 of [12] in that we lose a factor in our rates due to the hierarchical prior needing to be able to approximate the true parameter f0.”
Bayesian semiparametric stochastic volatility modeling (2010, methodology): “For example, when V1 = 1, Vj = 0 for j > 1, the SV-DPM model equals the autoregressive, stochastic volatility model of Jacquier et al. (1994). The SV-t model of Harvey et al.”
General maximum likelihood empirical Bayes estimation of normal means (2009, result): “Proof of Proposition 3. We provide a dense version of the proof since it is similar to the entropy calculations in [20, 21 , 39].”
Adaptive Bayesian density estimation with location-scale mixtures (2010, methodology): “The corresponding density is where We obtain a result similar to lemma 8 in [13].”