亚洲男人的天堂2018av,欧美草比,久久久久久免费视频精选,国色天香在线看免费,久久久久亚洲av成人片仓井空

While additional training data improves the robustness of deep neural networks against adversarial examples, it presents the challenge of curating a large number of specific real-world samples. We circumvent this challenge by using additional data from proxy distributions learned by state-of-the-art generative models. We first seek to formally understand the transfer of robustness from classifiers trained on proxy distributions to the real data distribution. We prove that the difference between the robustness of a classifier on the two distributions is upper bounded by the conditional Wasserstein distance between them. Motivated by our result, we next ask how to empirically select an appropriate generative model? We find that existing distance metrics, such as FID, fail to correctly determine the robustness transfer from proxy distributions. We propose a robust discrimination approach, which measures the distinguishability of synthetic and real samples under adversarial perturbations. Our approach accurately predicts the robustness transfer from different proxy distributions. After choosing a proxy distribution, the next question is which samples are most beneficial? We successfully optimize this selection by estimating the importance of each sample in robustness transfer. Finally, using our selection criterion for proxy distribution and individual samples, we curate a set of ten million most beneficial synthetic samples for robust training on the CIFAR-10 dataset. Using this set we improve robust accuracy by up to 7.5% and 6.7% in $\ell_{\infty}$ and $\ell_2$ threat model, and certified robust accuracy by 7.6% in $\ell_2$ threat model over baselines not using proxy distributions on the CIFAR-10 dataset.

The recursive and hierarchical structure of full rooted trees is used in various areas such as data compression, image processing, and machine learning. In most of these studies, the full rooted tree is not a random variable. It causes a problem of model selection to avoid overfitting. One method to solve it is to assume a prior distribution on the full rooted trees. It enables us to avoid overfitting based on the Bayes decision theory. For example, by assigning a low prior probability on a complex model, the MAP estimator prevents the overfitting. Further, we can avoid it by averaging all the models weighted by their posteriors. In this paper, we propose a probability distribution on a set of full rooted trees. Its parametric representation is well suited to calculate the properties of our distribution by recursive functions: the mode, the expectation, the posterior distribution, etc. Although some previous studies have proposed such distributions, they are for specific applications. Therefore, we extract the mathematically essential part of them and derive new generalized methods to calculate the expectation, the posterior distribution, etc.

Classical asymptotic theory for statistical inference usually involves calibrating a statistic by fixing the dimension $d$ while letting the sample size $n$ increase to infinity. Recently, much effort has been dedicated towards understanding how these methods behave in high-dimensional settings, where $d_n$ and $n$ both increase to infinity together at some prescribed relative rate. This often leads to different inference procedures, depending on the assumptions about the dimensionality, leaving the practitioner in a bind: given a dataset with 100 samples in 20 dimensions, should they calibrate by assuming $n \gg d$, or $d_n/n \approx 0.2$? This paper considers the goal of dimension-agnostic inference -- developing methods whose validity does not depend on any assumption on $d_n$. We introduce a new, generic approach that uses variational representations of existing test statistics along with sample splitting and self-normalization to produce a new test statistic with a Gaussian limiting distribution. The resulting statistic can be viewed as a careful modification of degenerate U-statistics, dropping diagonal blocks and retaining off-diagonals. We exemplify our technique for a handful of classical problems including one-sample mean and covariance testing. Our tests are shown to have minimax rate-optimal power against appropriate local alternatives, and without explicitly targeting the high-dimensional setting their power is optimal up to a $\sqrt 2$ factor. A hidden advantage is that our proofs are simple and transparent. We end by describing several fruitful open directions.

Bhattacharya (2021) has introduced a novel methodology for generating iid realizations from any target distribution on the Euclidean space, irrespective of dimensionality. In this article, our purpose is two-fold. We first extend the method for obtaining iid realizations from general multimodal distributions, and illustrate with a mixture of two 50-dimensional normal distributions. Then we extend the iid sampling method for fixed-dimensional distributions to variable-dimensional situations and illustrate with a variable-dimensional normal mixture modeling of the well-known "acidity data", with crucial application of the iid sampling method developed for multimodal distributions.

We obtain explicit $p$-Wasserstein distance error bounds between the distribution of the multi-parameter MLE and the multivariate normal distribution. Our general bounds are given for possibly high-dimensional, independent and identically distributed random vectors. Our general bounds are of the optimal $\mathcal{O}(n^{-1/2})$ order. Explicit numerical constants are given when $p\in(1,2]$, and in the case $p>2$ the bounds are explicit up to a constant factor that only depends on $p$. We apply our general bounds to derive Wasserstein distance error bounds for the multivariate normal approximation of the MLE in several settings; these being single-parameter exponential families, the normal distribution under canonical parametrisation, and the multivariate normal distribution under non-canonical parametrisation. In addition, we provide upper bounds with respect to the bounded Wasserstein distance when the MLE is implicitly defined.

We present a framework that allows for the non-asymptotic study of the $2$-Wasserstein distance between the invariant distribution of an ergodic stochastic differential equation and the distribution of its numerical approximation in the strongly log-concave case. This allows us to study in a unified way a number of different integrators proposed in the literature for the overdamped and underdamped Langevin dynamics. In addition, we analyse a novel splitting method for the underdamped Langevin dynamics which only requires one gradient evaluation per time step. Under an additional smoothness assumption on a $d$--dimensional strongly log-concave distribution with condition number $\kappa$, the algorithm is shown to produce with an $\mathcal{O}\big(\kappa^{5/4} d^{1/4}\epsilon^{-1/2} \big)$ complexity samples from a distribution that, in Wasserstein distance, is at most $\epsilon>0$ away from the target distribution.

We study distributionally robust optimization with Sinkorn distance -- a variant of Wasserstein distance based on entropic regularization. We derive convex programming dual reformulations when the nominal distribution is an empirical distribution and a general distribution, respectively. Compared with Wasserstein DRO, it is computationally tractable for a larger class of loss functions, and its worst-case distribution is more reasonable. To solve the dual reformulation, we propose an efficient batch gradient descent with a bisection search algorithm. Finally, we provide various numerical examples using both synthetic and real data to demonstrate its competitive performance.

The tensor Ising model is a discrete exponential family used for modeling binary data on networks with not just pairwise, but higher-order dependencies. In this exponential family, the sufficient statistic is a multi-linear form of degree $p\ge 2$, designed to capture $p$-fold interactions between the binary variables sitting on the nodes of a network. A particularly useful class of tensor Ising models are the tensor Curie-Weiss models, where one assumes that all $p$-tuples of nodes interact with the same intensity. Computing the maximum likelihood estimator (MLE) is computationally cumbersome in this model, due to the presence of an inexplicit normalizing constant in the likelihood, for which the standard alternative is to use the maximum pseudolikelihood estimator (MPLE). Both the MLE and the MPLE are consistent estimators of the natural parameter, provided the latter lies strictly above a certain threshold, which is slightly below $\log 2$, and approaches $\log 2$ as $p$ increases. In this paper, we compute the Bahadur efficiencies of the MLE and the MPLE above the threshold, and derive the optimal sample size (number of nodes) needed for either of these tests to achieve significance. We show that the optimal sample size for the MPLE and the MLE agree if either $p=2$ or the null parameter is greater than or equal to $\log 2$. On the other hand, if $p\ge 3$ and the null parameter lies strictly between the threshold and $\log 2$, then the two differ for sufficiently large values of the alternative. In particular, for every fixed alternative above the threshold, the Bahadur asymptotic relative efficiency of the MLE with respect to the MPLE goes to $\infty$ as the null parameter approaches the threshold. We also provide graphical presentations of the exact numerical values of the theoretical optimal sample sizes in different settings.

Adversarial training is among the most effective techniques to improve the robustness of models against adversarial perturbations. However, the full effect of this approach on models is not well understood. For example, while adversarial training can reduce the adversarial risk (prediction error against an adversary), it sometimes increase standard risk (generalization error when there is no adversary). Even more, such behavior is impacted by various elements of the learning problem, including the size and quality of training data, specific forms of adversarial perturbations in the input, model overparameterization, and adversary's power, among others. In this paper, we focus on \emph{distribution perturbing} adversary framework wherein the adversary can change the test distribution within a neighborhood of the training data distribution. The neighborhood is defined via Wasserstein distance between distributions and the radius of the neighborhood is a measure of adversary's manipulative power. We study the tradeoff between standard risk and adversarial risk and derive the Pareto-optimal tradeoff, achievable over specific classes of models, in the infinite data limit with features dimension kept fixed. We consider three learning settings: 1) Regression with the class of linear models; 2) Binary classification under the Gaussian mixtures data model, with the class of linear classifiers; 3) Regression with the class of random features model (which can be equivalently represented as two-layer neural network with random first-layer weights). We show that a tradeoff between standard and adversarial risk is manifested in all three settings. We further characterize the Pareto-optimal tradeoff curves and discuss how a variety of factors, such as features correlation, adversary's power or the width of two-layer neural network would affect this tradeoff.

Robust estimation is much more challenging in high dimensions than it is in one dimension: Most techniques either lead to intractable optimization problems or estimators that can tolerate only a tiny fraction of errors. Recent work in theoretical computer science has shown that, in appropriate distributional models, it is possible to robustly estimate the mean and covariance with polynomial time algorithms that can tolerate a constant fraction of corruptions, independent of the dimension. However, the sample and time complexity of these algorithms is prohibitively large for high-dimensional applications. In this work, we address both of these issues by establishing sample complexity bounds that are optimal, up to logarithmic factors, as well as giving various refinements that allow the algorithms to tolerate a much larger fraction of corruptions. Finally, we show on both synthetic and real data that our algorithms have state-of-the-art performance and suddenly make high-dimensional robust estimation a realistic possibility.