ME-Net defends neural nets against adversarial attacks by reconstructing images.
problem Adversarial attacks on deep neural networks.
method ME-Net uses matrix estimation to preprocess images, destroying adversarial noise while preserving global structure.
result ME-Net consistently outperforms state-of-the-art defenses on various benchmarks.
Generative Adversarial Nets [8] were recently introduced as a novel way to train generative models. In this work we introduce the conditional version of generative adversarial nets, which can be constructed by simply feeding the data, y, we wish to condition on to both the generator and discriminator. We show that this…
Improves neural net generalization by modeling hidden state distribution.
problem Brittleness and failure of existing neural nets, especially with sparse labeled data and adversarial training.
method State reification: modeling hidden state distribution and projecting test states towards it.
result Helps neural nets generalize better, especially with sparse labeled data and adversarial training.
Recent studies have highlighted the vulnerability of deep neural networks (DNNs) to adversarial examples - a visually indistinguishable adversarial image can easily be crafted to cause a well-trained model to misclassify. Existing methods for crafting adversarial examples are based on L2 and L∞ distortion me…
Our work proves convergence to low robust training loss for polynomial width ReLU networks.
problem Understanding why adversarial training leads to low robust training loss in over-parameterized neural nets.
method Extending convergence theory for standard supervised training to adversarial training, using tools from online learning and showing ReLU networks can approximate the step function.
result Convergence to low robust training loss for polynomial width ReLU networks under natural assumptions.
This paper offers a mathematical introduction to GANs.
problem Understanding GANs from a mathematical perspective.
method Mathematical analysis of GANs.
result Provides clarity for math-oriented students.
Generative model creates poisoning attacks against machine learning classifiers.
problem Poisoning attacks degrade machine learning classifier performance.
method Generative Adversarial Net with generator, discriminator, and classifier components.
result Effectively compromises machine learning classifiers, including deep networks.
Triple-GAIL learns from multiple sources to improve imitation learning for complex behaviors.
problem Limited scalability of GAIL in real-world scenarios like autonomous vehicles.
method Integrates expert demonstrations and generated experiences with an auxiliary skill selector.
result Triple-GAIL outperforms state-of-the-art methods in learning complex behaviors.
Adversarial training leads to clean data generalization with significant robust overfitting gap.
problem Significant robust generalization gap in adversarial training.
method Two theoretical views: representation complexity and training dynamics.
result ReLU nets with O(ND) extra parameters can achieve CGRO. Proposes KDA to protect deep nets from adversarial attacks.
problem Machine learning system vulnerability to adversarial attacks.
method Key based diversified aggregation with pre-filtering.
result Demonstrates high robustness and universality against various attacks.
We propose a general framework for increasing local stability of Artificial Neural Nets (ANNs) using Robust Optimization (RO). We achieve this through an alternating minimization-maximization procedure, in which the loss of the network is minimized over perturbed examples that are generated at each parameter update. We…
We propose a novel method for imputing missing data by adapting the well-known Generative Adversarial Nets (GAN) framework. Accordingly, we call our method Generative Adversarial Imputation Nets (GAIN). The generator (G) observes some components of a real data vector, imputes the missing components conditioned on what …
Adversarially-trained models transfer better in limited data scenarios.
problem Improving transfer learning performance with limited data.
method Adversarially-train deep nets, freeze early layers, fine-tune last layers, observe shape vs texture bias.
result Adversarially-trained models transfer better than non-adversarially-trained models, especially with limited data.
Deep generative models can learn to generate realistic-looking images, but many of the most effective methods are adversarial and involve a saddlepoint optimization, which requires a careful balancing of training between a generator network and a critic network. Maximum mean discrepancy networks (MMD-nets) avoid this i…
Paper finds the best way to estimate neural net distance from samples.
problem Estimating the neural net distance from samples.
method Developed minimax lower and upper bounds for the neural net distance.
result Lower and upper bounds match, validating the empirical neural net distance.
We show that training of generative adversarial network (GAN) may not have good generalization properties; e.g., training may appear successful but the trained distribution may be far from target distribution in standard metrics. However, generalization does occur for a weaker metric called neural net distance. It is a…
CGAN simulates time series data using categorical and continuous auxiliary info.
problem Simulating time series data with conditional information.
method Conditional Generative Adversarial Net (CGAN) for learning and generating time series data.
result CGAN can learn and generate various time series distributions and structures.
GANs have hidden assumptions causing practical issues.
problem GANs' practical performance contradicts theoretical guarantees.
method Proposed an alternative method to avoid these issues.
result Alternative method sidesteps the identified problems.
Symbolic knowledge in neural models can inadvertently make them more vulnerable to adversarial attacks.
problem Symbolic knowledge in neural models can make models more susceptible to adversarial attacks.
method Investigated deep probabilistic graphical models that incorporate symbolic knowledge and neural nets.
result Symbolic knowledge can propagate the negative effects of adversarial examples, making models more vulnerable.
Paper improves robustness and sparsity in adversarially trained DNNs.
problem Developing efficient compression algorithms for robustly trained DNNs.
method Pruning weights using relaxed augmented Lagrangian algorithms for both structured and unstructured levels, leveraging Feynman-Kac formalism.
result At least doubles channel sparsity of adversarially trained ResNet20 for CIFAR10 classification.
Class labels have been empirically shown useful in improving the sample quality of generative adversarial nets (GANs). In this paper, we mathematically study the properties of the current variants of GANs that make use of class label information. With class aware gradient and cross-entropy decomposition, we reveal how …
Study shows various pixel p-norm measures do not match human perception of adversarial attacks.
problem Understanding human perception of adversarial attacks on image classification systems.
method Performed a behavioral study comparing different p-norm measures and alternative metrics.
result Human perception of adversarial attacks does not align with pixel p-norm measures and other metrics.
Deterministic neural nets have been shown to learn effective predictors on a wide range of machine learning problems. However, as the standard approach is to train the network to minimize a prediction loss, the resultant model remains ignorant to its prediction confidence. Orthogonally to Bayesian neural nets that indi…
New SSIM measure improves adversarial attacks' perceptual quality.
problem Lack of perceptual quality in traditional adversarial attacks.
method Replaced Lp norms with SSIM measure for better quality. result SSIM-constrained attacks achieve better quality and similar success rate.
We provide an approach for learning deep neural net representations of models described via conditional moment restrictions. Conditional moment restrictions are widely used, as they are the language by which social scientists describe the assumptions they make to enable causal inference. We formulate the problem of est…
Study MinMax methods for optimization problems, including optimal transport.
problem Optimization problems, especially optimal transport.
method MinMax framework, regularization, neural networks, approximation theorems.
result Justification of neural networks for solving optimization problems.
DAL uses disentanglement for automatic labeling in GAN-based active learning.
problem Reducing human labeling in GAN-based active learning.
method DAL leverages disentanglement in InfoGAN to automatically label datapoints, deciding human labeling based on disagreement with InfoGAN labels and label correction.
result DAL achieves better performance than existing GAN-based active learning approaches on image classification tasks.
Improved DNN robustness to adversarial attacks using data-dependent activation and total variation minimization.
problem Improving Deep Neural Network robustness to adversarial attacks.
method Data-dependent activation function and total variation minimization.
result Robust accuracy of adversarially trained ResNet20 increased from ~46% to ~69% under IFGSM attack.
A machine learning approach optimizes data obfuscation for privacy.
problem Preserving data utility while protecting sensitive information.
method Generative adversarial networks (GAN) competition between generator and classifier.
result Optimal obfuscation mechanism reached through competitive nets.
Neural nets learn robust geometric data representations.
problem Ensuring neural networks are robust to adversarial attacks.
method Topological Data Analysis via persistence diagrams, Lipschitz stability.
result Certified ε-robustness on ORBIT5K dataset. The paper analyzes the statistical cost of tuning kernel hyperparameters in robust regression.
problem Finding the best interpolant from a class of kernels with unknown hyperparameters under adversarial noise.
method Finite-sample guarantees, subsampling guarantee for linear regression, ε-net argument for discretizing kernel parameterizations.
result Hyperparameter optimization increases sample complexity by just a logarithmic factor, compared to known parameters.
New method generates adversarial examples with conditional GAN.
problem Vulnerability of deep neural networks to adversarial examples.
method Conditional Generative Adversarial Network (cGAN) for generating adversarial examples.
result Our models reduce generating cost and improve robustness.
Generative model for inferring graph from time series data.
problem Generating graphs conditioned on multivariate time series data.
method Time Series Conditioned Graph Generation-Generative Adversarial Networks (TSGG-GAN).
result Demonstrates effectiveness and generalizability of TSGG-GAN on synthetic and real-world datasets.
The Madry Lab recently hosted a competition designed to test the robustness of their adversarially trained MNIST model. Attacks were constrained to perturb each pixel of the input image by a scaled maximal L∞ distortion ε = 0.3. This discourages the use of attacks which are not optimized on the L∞ dis…
Improved neural network robustness with instance-specific perturbation margins.
problem Adversarial training fails to generalize well to unperturbed test set.
method Instance adaptive adversarial training with sample-specific perturbation margins.
result Test accuracy improves with a marginal drop in robustness.
ZK-GanDef uses GANs to defend against adversarial examples without using them during training.
problem Adversarial examples challenge the effectiveness of neural network classifiers.
method Generative adversarial net (GAN) based zero knowledge adversarial training defense.
result Enhances test accuracy on adversarial examples by up to 49.17% compared to zero knowledge approaches.
Generative modeling over natural images is one of the most fundamental machine learning problems. However, few modern generative models, including Wasserstein Generative Adversarial Nets (WGANs), are studied on manifold-valued images that are frequently encountered in real-world applications. To fill the gap, this pape…
Adversarial attacks can fool copyright detection systems.
problem Vulnerability of copyright detection systems to adversarial attacks.
method Used gradient methods to create adversarial music that fooled detection systems.
result Adversarial attacks can successfully deceive industrial copyright detection tools.
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.
RODE-Net learns ODEs from data with random parameters using neural networks and GANs.
problem Learning ODEs from data with unknown and random parameters.
method RODE-Net combines symbolic networks and GANs to estimate both the ODE and its parameters.
result RODE-Net can accurately estimate the distribution of model parameters and make reliable predictions.
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…
GanDef uses GANs to defend against adversarial examples in neural networks.
problem Defending against adversarial examples in neural networks.
method GAN-based adversarial training defense using a competition game to regulate feature selection.
result GanDef trains a classifier to defend against adversarial examples with high accuracy.
HexaGAN tackles real-world classification issues with missing data, class imbalance, and missing labels.
problem Missing data, class imbalance, and missing labels in real-world data.
method Generative adversarial network framework with six components and novel loss functions.
result Up to 5% improvement in classification performance compared to state-of-the-art methods.
We present Optimal Transport GAN (OT-GAN), a variant of generative adversarial nets minimizing a new metric measuring the distance between the generator distribution and the data distribution. This metric, which we call mini-batch energy distance, combines optimal transport in primal form with an energy distance define…
New method uses GANs and proper scoring rules for robust scatter estimation.
problem Robust scatter estimation in statistics.
method General learning via classification framework based on proper scoring rules.
result Proposed robust scatter estimators achieve minimax rate under Huber's contamination model.
Recently, a unified model for image-to-image translation tasks within adversarial learning framework has aroused widespread research interests in computer vision practitioners. Their reported empirical success however lacks solid theoretical interpretations for its inherent mechanism. In this paper, we reformulate thei…
The goal of this paper is to analyze the geometric properties of deep neural network classifiers in the input space. We specifically study the topology of classification regions created by deep networks, as well as their associated decision boundary. Through a systematic empirical investigation, we show that state-of-t…
Develops asymptotic theory for adversarial estimators.
problem Estimating unknown functions in machine learning and econometrics.
method Derives convergence rates and normality of A-estimators under various conditions.
result Normality of neural-net M-estimators, overcoming previous technical issues.