The paper proves polynomial mixing time upper bounds for three data augmentation algorithms (ProbitDA, LogitDA, LassoDA) with explicit dependence on design matrix, prior, n, and d.
Spectral gaps, symmetries and log-concave perturbations
1 Pith paper cite this work. Polarity classification is still indexing.
abstract
We discuss situations where perturbing a probability measure on $\mathbb{R}^n$ does not deteriorate its Poincar\'e constant by much. A particular example is the symmetric exponential measure in $\mathbb{R}^n$, even log-concave perturbations of which have Poincar\'e constants that grow at most logarithmically with the dimension. This leads to estimates for the Poincar\'e constants of $(n/2)$-dimensional sections of the unit ball of $\ell_p^n$ for $1 \leq p \leq 2$, which are optimal up to logarithmic factors. We also consider symmetry properties of the eigenspace of the Laplace-type operator associated with a log-concave measure. Under symmetry assumptions we show that the dimension of this space is exactly $n$, and we exhibit a certain interlacing between the "odd" and "even" parts of the spectrum.
fields
math.ST 1years
2024 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Fast Mixing of Data Augmentation Algorithms: Bayesian Probit, Logit, and Lasso Regression
The paper proves polynomial mixing time upper bounds for three data augmentation algorithms (ProbitDA, LogitDA, LassoDA) with explicit dependence on design matrix, prior, n, and d.