Paper develops robust Bayesian models for linear regression under adversarial perturbations.
problem Ensuring reliable machine learning models under data perturbations.
method Formulates adversarial Bregman divergence loss, computes adversarial perturbation, introduces adversarially robust posteriors, derives generalization certificates.
result Derives first rigorous generalization certificates for adversarially robust Bayesian linear regression.
New method for bidirectional generative modeling using adversarial gradient estimation.
problem Bidirectional generative modeling with various f-divergences. method Adversarial gradient estimation for f-divergence optimization. result Similar algorithms for different f-divergences with varying scaling. Unified ML and adversarial learning via α-divergence.
problem Combining strengths of ML and adversarial learning for better generative models.
method Proposes an α-Bridge to unify ML and adversarial learning using α-divergence.
result Generalizations of the α-Bridge are related to recent adversarial learning regularization approaches.
New principle controls graph-informed adversarial discrepancies.
problem Graph-informed adversarial learning for interpolative divergences.
method Proves infimal subadditivity for interpolative divergences.
result Graph-informed adversarial learning is justified for interpolative divergences.
Proposes ATM method to improve domain adaptation.
problem Mitigating distribution divergence between source and target domains.
method Adversarial Tight Match (ATM) method using Maximum Density Divergence (MDD).
result New state-of-the-art performance on domain adaptation benchmarks.
Parametric adversarial divergences, which are a generalization of the losses used to train generative adversarial networks (GANs), have often been described as being approximations of their nonparametric counterparts, such as the Jensen-Shannon divergence, which can be derived under the so-called optimal discriminator …
New model learns better policies from expert demonstrations with higher efficiency.
problem Learning accurate policies from expert demonstrations with high efficiency.
method Generative adversarial imitation learning (GAIL) model that learns f-divergence automatically. result Learns better policies with higher data efficiency in physics-based control tasks.
Study statistical guarantees for DRO with OT and OT-regularized divergences.
problem Enhancing adversarial robustness in machine learning models.
method Derive concentration inequalities for supervised learning via DRO-based adversarial training.
result First to cover soft-constraint costs and reweighting mechanisms in adversarial training.
New algorithms improve robust estimation in contaminated Gaussian models.
problem Simultaneous estimation of location and variance matrix in contaminated Gaussian models.
method Tractable adversarial algorithms with spline discriminators for robust estimation.
result Achieve minimax optimal rates or near-optimal rates under Huber's contamination model.
New method enhances adversarial robustness of deep learning models.
problem Improving the robustness of deep learning models against adversarial attacks.
method Optimal transport regularized divergences applied to distributionally robust optimization.
result Improved adversarial robustness on CIFAR-10 and CIFAR-100 datasets.
New framework using Jensen-Shannon divergence improves domain adaptation theory.
problem Incoherence between empirical domain adversarial training and theoretical H-divergence. method Established new theoretical framework based on Jensen-Shannon divergence, derived bi-directional upper bounds.
result Framework exhibits flexibilities for various transfer learning problems.
Paper explores how generative models can be made more creative.
problem Limitation of generative models in diverging from original data distribution.
method Proposes a novel training objective called Bounded Adversarial Divergence (BAD) to enable creative divergence.
result Preliminary results suggest BAD can enable creative divergence in generative models.
Unified framework improves neural network robustness against label noise and adversarial attacks.
problem High sensitivity of neural networks to data contamination, including label noises and adversarial perturbations.
method Unified minimum-divergence estimation problem, rSDNet framework.
result Improves robustness to label corruption and adversarial attacks while maintaining competitive accuracy on clean data.
QUAM improves uncertainty quantification in deep learning models.
problem Estimating epistemic uncertainty in deep learning models.
method QUAM identifies regions with high divergence between predictions and a reference model.
result QUAM has lower approximation error of epistemic uncertainty compared to previous methods.
Generative adversarial networks (GAN) approximate a target data distribution by jointly optimizing an objective function through a "two-player game" between a generator and a discriminator. Despite their empirical success, however, two very basic questions on how well they can approximate the target distribution remain…
MonoFlow rethinks GANs using Wasserstein gradient flows.
problem Inconsistencies between GAN theory and practice.
method Unified generative modeling framework based on Wasserstein gradient flows.
result Adversarial training can be seen as particle flow optimization.
New method improves imitation learning from expert observations.
problem Challenges in imitation learning from observation setting.
method Reparameterized Variational Divergence Minimization.
result Our method outperforms baseline approaches in low-dimensional tasks.
Recent years have seen adversarial losses been applied to many fields. Their applications extend beyond the originally proposed generative modeling to conditional generative and discriminative settings. While prior work has proposed various output activation functions and regularization approaches, some open questions …
Generative neural samplers are probabilistic models that implement sampling using feedforward neural networks: they take a random input vector and produce a sample from a probability distribution defined by the network weights. These models are expressive and allow efficient computation of samples and derivatives, but …
Generalized dual discriminator GANs improve upon traditional GANs by using two discriminators and a flexible loss function.
problem Mode collapse in GANs.
method Introducing dual discriminator α-GANs and extending the approach to arbitrary functions. result The approach reduces the optimization problem to a linear combination of an f-divergence and a reverse f-divergence. Improves deep learning robustness by enforcing local and global compactness.
problem Deep neural networks' vulnerability to adversarial attacks.
method Proposes Adversary Divergence Reduction Network (ADRN) that enforces local/global compactness and clustering assumption.
result Augmenting adversarial training with ADRN components improves robustness.
The paper analyzes the statistical properties of GANs using f-divergence.
problem Understanding the statistical behavior of GANs and comparing different f-divergences. method Asymptotic analysis of f-divergence GANs, including Kullback-Leibler divergence. result Asymptotically equivalent GANs with the same discriminator classes for correctly specified models.
ANTIDOTE reduces noisy labels influence during learning.
problem Learning with noisy labels.
method Information-divergence neighborhood relaxation and adversarial training.
result ANTIDOTE outperforms standard cross-entropy loss in noisy label settings.
Adversarial Optimization (AO) provides a reliable, practical way to match two implicitly defined distributions, one of which is usually represented by a sample of real data, and the other is defined by a generator. Typically, AO involves training of a high-capacity model on each step of the optimization. In this work, …
CD learning is shown to be an adversarial game for fitting models.
problem Difficulty in understanding the convergence properties of CD learning.
method Presented an alternative derivation of CD without approximation, showing it as a time-reversal adversarial game.
result CD is an adversarial learning procedure where a discriminator tries to classify time-reversed Markov chains.
Generative adversarial networks (GANs) are successful deep generative models. GANs are based on a two-player minimax game. However, the objective function derived in the original motivation is changed to obtain stronger gradients when learning the generator. We propose a novel algorithm that repeats the density ratio e…
New theory explains GAN's high quality but low diversity.
problem Lack of theoretical justification for non-saturating GAN training.
method Showed non-saturating GAN training approximately minimizes a specific f-divergence.
result Non-saturating GAN training minimizes a particular f-divergence.
The paper improves model robustness by regularizing posterior differences.
problem Improving model robustness in noisy input scenarios.
method Posterior differential regularization with f-divergence. result Regularizing with f-divergence improves model robustness. To obtain uncertainty estimates with real-world Bayesian deep learning models, practical inference approximations are needed. Dropout variational inference (VI) for example has been used for machine vision and medical applications, but VI can severely underestimates model uncertainty. Alpha-divergences are alternative …
Paper proposes f-EBM for training deep EBMs using various f-divergences.
problem Training deep EBMs with intractable partition functions.
method Introduces f-EBM framework and optimization algorithm for any f-divergence.
result f-EBM outperforms contrastive divergence and other f-divergences.
A new method improves adversarial robustness by optimizing importance weights.
problem Adversarial training's non-uniform robustness across different data points.
method Doubly-robust instance reweighted adversarial training using distributionally robust optimization.
result Improves robustness against attacks on the weakest data points.
WDAIL uses Wasserstein distance for more effective reward shaping in IL.
problem Fixed reward functions in GAIL limit performance on complex tasks.
method Introduces Wasserstein distance and PPO for improved reward shaping and stability.
result Significant performance improvement in complex MuJoCo tasks.
Probabilistic models are often trained by maximum likelihood, which corresponds to minimizing a specific f-divergence between the model and data distribution. In light of recent successes in training Generative Adversarial Networks, alternative non-likelihood training criteria have been proposed. Whilst not necessarily…
Improved GANs estimate convergence rate for density estimation.
problem Improving the accuracy of density estimation with GANs.
method Proved an oracle inequality for JS divergence between GAN estimate and true density.
result JS-divergence rate of convergence is (logn/n)2β/(2β+d). New method improves adversarial robustness of neural networks.
problem Vulnerability of neural networks to adversarial attacks.
method Integrates φ-divergence regularization into distributionally robust risk function.
result Achieves near-optimal sample complexity with efficient stochastic gradient methods.
A new form of the variational autoencoder (VAE) is proposed, based on the symmetric Kullback-Leibler divergence. It is demonstrated that learning of the resulting symmetric VAE (sVAE) has close connections to previously developed adversarial-learning methods. This relationship helps unify the previously distinct techni…
A new method detects and corrects adversarial attacks on classifiers.
problem Detecting and correcting adversarial attacks on machine learning models.
method Unsupervised autoencoder trained with KL divergence loss to match predictions.
result Almost completely neutralizes powerful attacks on MNIST and Fashion-MNIST.
This paper proposes the divergence triangle as a framework for joint training of generator model, energy-based model and inference model. The divergence triangle is a compact and symmetric (anti-symmetric) objective function that seamlessly integrates variational learning, adversarial learning, wake-sleep algorithm, an…
Generative adversarial networks (GANs) can be interpreted as an adversarial game between two players, a discriminator D and a generator G, in which D learns to classify real from fake data and G learns to generate realistic data by "fooling" D into thinking that fake data is actually real data. Currently, a dominating …
Develops a new divergence framework that combines f-divergences and IPMs.
problem Comparing distributions that are not absolutely continuous.
method Introduces (f,Γ)-divergences as a two-stage mass-redistribution/mass-transport process. result Improves estimation, learning, and uncertainty quantification in GANs for heavy-tailed distributions.
Paper tackles continuous transfer learning with evolving target domains.
problem Challenges of negative transfer in evolving target domains.
method Proposes label-informed C-divergence for measuring distribution shift and negative transfer.
result Demonstrates effectiveness of TransLATE framework in minimizing classification error and C-divergence.
In this note, we point out a basic link between generative adversarial (GA) training and binary classification -- any powerful discriminator essentially computes an (f-)divergence between real and generated samples. The result, repeatedly re-derived in decision theory, has implications for GA Networks (GANs), providing…
Adversarial perturbations dramatically decrease the accuracy of state-of-the-art image classifiers. In this paper, we propose and analyze a simple and computationally efficient defense strategy: inject random Gaussian noise, discretize each pixel, and then feed the result into any pre-trained classifier. Theoretically,…
Introduces Cauchy-Schwarz divergence for domain adaptation.
problem Evaluating discrepancy between source and target domains in unsupervised domain adaptation.
method Introduces Cauchy-Schwarz divergence as a measure for evaluating discrepancy between marginal and conditional distributions.
result CS divergence offers a tighter generalization error bound than Kullback-Leibler divergence.
This short note highlights some links between two lines of research within the emerging topic of trustworthy machine learning: differential privacy and robustness to adversarial examples. By abstracting the definitions of both notions, we show that they build upon the same theoretical ground and hence results obtained …
This paper raises an implicit manifold learning perspective in Generative Adversarial Networks (GANs), by studying how the support of the learned distribution, modelled as a submanifold Mθ, perfectly match with Mr, the support of the real data distribution. We show that optimizing Jensen-Sha…
A new, computationally friendly formula for a class of risk-averse preferences.
problem Characterizing a class of risk-averse preferences called uniformly weighted divergence preferences.
method Introducing a new formula that characterizes UWDP as the translation-invariant hull of state-independent expected utility.
result UWDP are the translation-invariant hull of state-independent expected utility over L0. DART tackles adversarial robustness in domain adaptation without labeled target data.
problem Combining distribution shifts and adversarial examples in unsupervised domain adaptation.
method DART framework that combines standard UDA methods with a generalization bound for adversarial target loss.
result DART significantly enhances model robustness to adversarial attacks compared to state-of-the-art methods.