Adversarial training improves robustness against common corruptions.
problem Improving robustness of models against common corruptions.
method Adversarial training with ℓ p \ell_p ℓ p perturbation and learned perceptual similarity. result Our approach leads to state-of-the-art performance on common corruptions.
Study linear regression with missing or corrupted data, showing error bounds.
problem Linear regression under missing or corrupted data.
method Information-theoretic lower bounds and efficient algorithms.
result Error bounds match in missing and corruption settings.
RILA learns HQMMs robustly against adversarial corruption.
problem Robustness of HQMM learning algorithms under adversarial perturbations.
method Adversarially Corrupted HQMM (AC-HQMM) and Robust Iterative Learning Algorithm (RILA).
result RILA outperforms existing algorithms in convergence stability, corruption resilience, and physical validity.
Monotone adversarial corruptions degrade optimal learning algorithms.
problem Optimal learning algorithms' reliance on exchangeability and independence is challenged.
method Introduces a monotone adversarial corruption model where an adversary adds monotone corruptions to a clean dataset.
result Optimal learning algorithms achieve suboptimal expected error on new test points.
New algorithm learns reliable regression coefficients from streaming data with partial features and adversarial corruption.
problem Learning reliable regression coefficients from streaming data with partial features and adversarial corruption.
method RoOFS algorithm that iteratively updates regression coefficients and uncorrupted feature set via robust online feature substitution.
result RoOFS algorithm has a restricted error bound compared to the optimal solution and outperforms existing methods in feature selection and regression coefficient recovery.
New subspace methods resist limited adversary corruption on test instances.
problem Adversary can corrupt up to l features in test instances.
method Subspace voting techniques to transform algorithms.
result Significant fraction of voting hypotheses do not contain corrupt features.
Improved SGD for robust linear and ReLU regression with adversarial corruptions.
problem Robust regression with adversarial corruptions in streaming data.
method Stochastic gradient descent (SGD-exp) with exponentially decaying step size.
result Nearly linear convergence to true parameter with up to 50% Massart corruption rate.
New algorithm for linear optimization with adaptive corruption.
problem Stochastic linear optimization under adversarial corruption.
method Algorithm uses Löwner-John's ellipsoid for exploration and divides time into epochs.
result Regret increases linearly with corruption amount.
New algorithm reduces regret in corrupted bandits.
problem Stochastic multi-armed bandits with adversarial corruption.
method A new algorithm that is agnostic to corruption levels.
result Regret is nearly optimal and can handle significant corruption.
We introduce a new model of stochastic bandits with adversarial corruptions which aims to capture settings where most of the input follows a stochastic pattern but some fraction of it can be adversarially changed to trick the algorithm, e.g., click fraud, fake reviews and email spam. The goal of this model is to encour…
Proposes a robust IV estimator using optimal transport for corrupted or adversarial data.
problem Lack of robustness in traditional IV estimators for corrupted or adversarial data.
method Integrates data-derivative information through optimal transport to address geometric aspects of data.
result Improves robustness against data corruption and adversarial attacks.
Unified framework SVAM learns GLMs robustly to adversarial label corruption.
problem Learning GLMs under adversarial label corruption.
method SVAM framework based on variance reduction technique.
result Provable model recovery guarantees superior to state-of-the-art.
Study improves image classifier robustness to random p-norm corruptions.
problem Improving robustness of image classifiers to real-world imperceptible corruptions.
method Training and testing with random p-norm corruptions, evaluating robustness against different p-norms.
result Training with a combination of p-norm corruptions significantly improves robustness.
Algorithm maximizes total reward in multi-agent bandits with adversarial corruptions.
problem Maximizing total reward in multi-agent bandits with adversarial corruptions.
method Proposes a cooperative learning algorithm robust to adversarial corruptions.
result Demonstrates an additive O ( ( L / L min ) C ) O((L / L_{\min}) C) O (( L / L m i n ) C ) regret term for an adversary with unknown corruption budget. New algorithm resists corruption in linear contextual bandits.
problem Adversarial corruption in linear contextual bandits.
method Variance-aware algorithm with multi-level partition and adaptive confidence sets.
result Regret bound of i l d e O ( C 2 d ∑ t = 1 T σ t 2 + C 2 R d T ) ilde{O}(C^2d\sqrt{\sum_{t = 1}^T σ_t^2} + C^2R\sqrt{dT}) i l d e O ( C 2 d ∑ t = 1 T σ t 2 + C 2 R d T ) . Robustly estimates mean with quantized data and corruption.
problem Mean estimation under quantization and adversarial corruption.
method Constructs multivariate robust estimators in two settings.
result Optimal estimators up to logarithmic factors.
New algorithm tackles adversarial corruption in Lipschitz bandits with sub-linear regret.
problem Adversarial corruption in Lipschitz bandits.
method Developed robust Lipschitz bandit algorithms for weak and strong adversaries.
result Achieved sub-linear regret under both weak and strong adversaries.
Study shows multi-source learning is more resilient to adversarial corruption than single-source learning.
problem Learning from multiple untrusted data sources, especially when some are adversarially corrupted.
method Analyzed the scenario where an adversary can corrupt a fixed fraction of data sources, derived a generalization bound for this setting.
result PAC-learnability is possible in the multi-source setting even when some data sources are adversarially corrupted.
New study shows how adversaries can bias fair machine learning models even with corrupted data.
problem Fairness concerns in machine learning models under data corruption.
method Study of fairness-aware learning algorithms under worst-case data manipulations.
result Natural learning algorithms optimizing for both accuracy and fairness are order-optimal in terms of corruption ratio and protected groups frequencies.
SGD-trained neural networks generalize well even with adversarial label noise.
problem Generalization of neural networks trained on adversarial label noise.
method Training a one-hidden-layer neural network with SGD on arbitrary width networks.
result SGD-trained networks achieve classification accuracy competitive with the best halfspace over adversarial label noise.
New method tackles adversarial sign-corrupted isotonic regression, estimating monotonic signals under heavy dependence.
problem Estimating monotonic signals when responses are sign-corrupted and adversarially designed to violate monotonicity.
method Developed ASCIFIT, a three-step estimation procedure using PAVA with pre- and post-processing corrections.
result Theoretical guarantees of sharp high probability upper bounds and minimax lower bounds for ASCIFIT.
New algorithm reduces linear contextual bandit regret with adversarial corruption.
problem Linear contextual bandit with adversarial reward corruption.
method Optimism in the face of uncertainty principle, weighted ridge regression.
result Achieves nearly optimal regret for both corrupted and uncorrupted cases.
Analyzes biased random walks and corrupted intervals in adversarial settings.
problem Learning thresholds and intervals in adversarial conditions.
method Analyzes biased random walks and corrupted intervals under adversarial design.
result Analyzes the expected behavior of biased random walks and corrupted intervals.
New algorithm detects communities even with corrupted data, reaching Kesten-Stigum threshold.
problem Robust community detection in stochastic block model with node corruptions.
method Polynomial-time algorithm using Grothendieck norm of principal submatrices.
result First algorithm to achieve weak recovery at Kesten-Stigum threshold with node corruptions.
Class selectivity affects robustness to corruptions but not to adversarial attacks.
problem Understanding the relationship between class selectivity and robustness in neural networks.
method Investigated the impact of class selectivity on robustness to natural corruptions and adversarial attacks using Tiny ImageNetC and CIFAR10C datasets.
result Decreasing class selectivity increases robustness to both natural corruptions and adversarial attacks.
A new algorithm reduces regret in bandit problems with adversarial corruptions.
problem Optimizing decision-making in bandit problems with variable uncertainties and adversarial interference.
method Proposes HCW-GLB-OMD, an OMD-based estimator with Hessian-based confidence weights for robustness.
result Achieves instance-wise minimax optimality with a κ κ κ -factor in the corruption term. New algorithm finds corrupted vertices in graphs with few queries.
problem Adversarial tampering of graph edges and vertices.
method Active learning algorithm with polynomial query complexity.
result Efficiently recovers corrupted vertices with small query complexity.
New method improves neural network robustness without sacrificing generalization.
problem Robustness and generalization are often at odds in neural networks.
method Distributionally robust loss function bridging robustness and generalization.
result Certified robustness against data evasion and poisoning attacks with guaranteed generalization.
Solving inverse problems continues to be a central challenge in computer vision. Existing techniques either explicitly construct an inverse mapping using prior knowledge about the corruption, or learn the inverse directly using a large collection of examples. However, in practice, the nature of corruption may be unknow…
This study tackles adversarial corruption in model-based reinforcement learning.
problem Adversarial corruption in model-based reinforcement learning.
method Maximum likelihood estimation (MLE) approach for learning transition model in both online and offline settings.
result Proves a regret of i l d e O ( T + C ) ilde{\mathcal{O}}(\sqrt{T} + C) i l d e O ( T + C ) for CR-OMLE and a suboptimality of O ( C / n ) \mathcal{O}(C/n) O ( C / n ) for CR-PMLE. The paper shows optimal robustness against adversarial corruption in sequential decision-making problems.
problem Optimal robustness to adversarial corruption in online decision-making problems.
method Investigates prediction with expert advice and multi-armed bandit problems, focusing on algorithms with decreasing learning rates and second-order regret bounds.
result Optimal robustness can be expressed by a square-root dependency on the amount of corruption, achieving O ( log N Δ + C log N Δ ) O(\frac{\log N}{\Delta} + \sqrt{\frac{C \log N}{\Delta}}) O ( Δ l o g N + Δ C l o g N ) -regret. Robust algorithm optimizes corrupted Gaussian process bandits.
problem Sequential optimization of corrupted, expensive reward functions.
method Robust GP Phased Elimination (RGP-PE) algorithm.
result Algorithm balances robustness to corruptions with exploration and exploitation.
New algorithm robust to label corruptions in active learning.
problem Active learning under unknown adversarial label corruptions.
method Proposed a new active learning algorithm that is provably correct without assumptions on corruptions.
result Achieves minimax label complexity in non-corrupted setting and only requires additional labels to achieve desired accuracy in corrupted setting.
PDA improves deep neural networks' robustness against adversarial and common corruptions.
problem Deep neural networks' lack of robustness against common corruptions and adversarial attacks.
method Progressive Data Augmentation (PDA) that injects diverse adversarial noises during training.
result PDA-trained networks are more robust against both adversarial and common corruptions.
Adversarial examples and noisy images share a common cause.
problem Improving machine learning models' robustness to adversarial attacks and random noise.
method Empirical and theoretical analysis of adversarial examples and corrupted images.
result Adversarial robustness and corruption robustness are closely related.
Efficiently recovers data corrupted by adversarial noise in structured settings.
problem Recovering clean data points from corrupted Gaussian data with low-rank noise and adversarial coordinate corruptions.
method Developed an efficient algorithm using a combinatorial approach to analyze Basis Pursuit (BP) method.
result Achieved nearly-optimal recovery of data points up to a i l d e O ( k s / d ) ilde O(ks/d) i l d e O ( k s / d ) error bound. Improved regret bounds for Tsallis-INF in adversarial bandits and corruptions.
problem Adversarial bandits and corruptions in multiarmed bandit problems.
method Improved regret bounds for Tsallis-INF algorithm.
result Achieves $\mathcal{O}\left(\left(\sum_{i
eq i^*} \frac{1}{Δ_i}
ight)\log_+\left(\frac{(K-1)T}{\left(\sum_{i
eq i^*} \frac{1}{Δ_i}
ight)^2}
ight)+\sqrt{C\left(\sum_{i
eq i^*}\frac{1}{Δ_i}
ight)\log_+\left(\frac{(K-1)T}{C\sum_{i
eq i^*}\frac{1}{Δ_i}}
ight)}
ight)$ regret bound.
New algorithms improve contextual search in the presence of adversarial corruptions.
problem Improving search accuracy in dynamic pricing settings with corrupted responses.
method Two algorithms based on binary search and gradient descent methods.
result Achieve near-optimal regret in the absence of adversarial corruptions and gracefully degrade with corrupted agents.
New method for robust learning from batches, even adversarial ones.
problem Learning from batches that may be corrupt or adversarial.
method General framework for robust learning, derived from optimal robust algorithms.
result First robust agnostic learning algorithms for various distributions.
Improved ε ε ε -greedy handles strategic bidding in PPC auctions.
problem Strategic bidding in PPC auctions with personalization and corruptions.
method Extended ε ε ε -greedy to handle strategic arms in contextual multi-arm bandit. result ε ε ε -greedy is robust to adversarial corruptions and degrades linearly with corruption. New method solves group synchronization with cycle-edge message passing.
problem Solving group synchronization with adversarial or uniform corruption and small noise.
method Cycle-edge message passing procedure using cycle consistency information.
result Exact recovery and linear convergence guarantees under adversarial corruption.
Estimates GLMs robustly against label corruptions.
problem Learning GLMs under adversarial label corruptions.
method Iterative trimmed maximum likelihood estimator.
result Achieves minimax near-optimal risk.
Enhances neural network robustness with Mixup and TLAT.
problem Neural networks are sensitive to various perturbations and adversarial examples.
method Combines Mixup augmentation with Targeted Labeling Adversarial Training (TLAT).
result M-TLAT increases robustness against 19 corruptions and 5 adversarial attacks without reducing clean sample accuracy.
New algorithm for robust density estimation in corrupted data.
problem Density estimation in the presence of adversarial corruption.
method Proposes an algorithm for constructing a density estimator within a star-shaped density class, derived minimax bounds for estimation.
result Obtained minimax upper and lower bounds for density estimation under adversarial corruption.
Unified framework for corruption-robust linear bandits with optimal gap-dependent misspecification bounds.
problem Effective learning in linear bandits with corrupted rewards across different corruption models.
method Unified framework for analyzing strong and weak corruption, connection to gap-dependent misspecification, and specialized algorithm.
result Optimal bounds for gap-dependent misspecification in linear bandits.
This paper benchmarks neural network robustness to corruptions and perturbations.
problem Establishing benchmarks for image classifier robustness to corruptions and perturbations.
method Developed ImageNet-C and ImageNet-P datasets to evaluate robustness to corruptions and perturbations, not adversarial attacks.
result There are negligible changes in relative corruption robustness from AlexNet to ResNet classifiers.
Picket guards against corrupted data in machine learning models.
problem Data corruption biases models and invalidates predictions.
method PicketNet detects corrupted data using self-supervised deep learning; flags corrupted queries online.
result Picket consistently protects models from corrupted data during training and deployment.
Model predicts counterfactuals under domain shift and inaccessible variables.
problem Runtime domain corruption impairs counterfactual prediction.
method Subsumes counterfactual prediction under domain adaptation, uses adversarial domain adaptation to reduce distribution disparity.
result VEGAN outperforms baselines in individual-level treatment effect estimation.