Study improves resilience against adversarial clean-label attacks in real and noisy settings.
problem Ensuring accurate predictions in the presence of adversarial clean-label samples.
method Sequential learning from a stream of i.i.d. data, allowing abstention for uncertain predictions.
result Theoretical analysis and adaptations for the agnostic setting with a clean-label adversary and noise.
Bullseye Polytope improves clean-label poisoning attacks in transfer learning.
problem Poisoning neural networks with correctly labeled data.
method Creates poison images with centers close to target images in feature space.
result Improves attack success rate by 26.75% in end-to-end transfer learning.
SSMs can be poisoned with clean labels, leading to generalization failure.
problem The implicit bias of SSMs can be manipulated by including special training examples with clean labels.
method Formal proof and empirical demonstration of the phenomenon.
result SSMs can fail to generalize even with clean labels, due to the inclusion of special training examples.
INN method refines clean labeled data from noisy labels.
problem Handling noisy labels in deep neural networks.
method INN method based on memorization effect at neighbor regions.
result INN method resolves memorization effect shortcomings.
MetaPoison poisons neural networks by learning to craft imperceptible changes to training data.
problem Data poisoning attacks on neural networks.
method MetaPoison uses meta-learning to approximate bilevel optimization for crafting poisons.
result MetaPoison outperforms previous methods and works in various scenarios.
In unsupervised domain adaptation (UDA), classifiers for the target domain (TD) are trained with clean labeled data from the source domain (SD) and unlabeled data from TD. However, in the wild, it is difficult to acquire a large amount of perfectly clean labeled data in SD given limited budget. Hence, we consider a new…
Clean-label poisoning attacks inject innocuous looking (and "correctly" labeled) poison images into training data, causing a model to misclassify a targeted image after being trained on this data. We consider transferable poisoning attacks that succeed without access to the victim network's outputs, architecture, or (i…
Collecting large training datasets, annotated with high-quality labels, is costly and time-consuming. This paper proposes a novel framework for training deep convolutional neural networks from noisy labeled datasets that can be obtained cheaply. The problem is formulated using an undirected graphical model that represe…
We consider the learning from noisy labels (NL) problem which emerges in many real-world applications. In addition to the widely-studied synthetic noise in the NL literature, we also consider the pseudo labels in semi-supervised learning (Semi-SL) as a special case of NL. For both types of noise, we argue that the gene…
Deep learning with noisy labels is practically challenging, as the capacity of deep models is so high that they can totally memorize these noisy labels sooner or later during training. Nonetheless, recent studies on the memorization effects of deep neural networks show that they would first memorize training data of cl…
This work explores how neural network architecture affects robustness to noisy labels.
problem The impact of neural network architecture on robustness to noisy labels.
method Formal framework connecting robustness to architecture alignments, measured by predictive power in representations.
result Network robustness to noisy labels improves when its architecture is more aligned with the target function.
Multi-domain image-to-image translation is a problem where the goal is to learn mappings among multiple domains. This problem is challenging in terms of scalability because it requires the learning of numerous mappings, the number of which increases proportional to the number of domains. However, generative adversarial…
Data poisoning is an attack on machine learning models wherein the attacker adds examples to the training set to manipulate the behavior of the model at test time. This paper explores poisoning attacks on neural nets. The proposed attacks use "clean-labels"; they don't require the attacker to have any control over the …
Framework prevents deep learning models from memorizing noisy labels.
problem Deep learning models memorize noisy labels during early learning phase.
method Develops a technique that exploits early learning phase via regularization.
result Framework achieves robustness to noisy annotations on benchmarks and real-world datasets.
In this paper we are interested in the prediction of preterm birth based on diagnosis codes from longitudinal EHR. We formulate the prediction problem as a supervised classification with noisy labels. Our base classifier is a Recurrent Neural Network with an attention mechanism. We assume the availability of a data sub…
A new method improves graph-based semi-supervised classification by removing noise and mixed signs.
problem Inaccurate soft labels and noise in graph-based semi-supervised classification.
method Triple-matrix-recovery-based robust auto-weighted label propagation framework (ALP-TMR).
result Improved robustness to noise and outliers in label estimation.
Datasets with significant proportions of noisy (incorrect) class labels present challenges for training accurate Deep Neural Networks (DNNs). We propose a new perspective for understanding DNN generalization for such datasets, by investigating the dimensionality of the deep representation subspace of training samples. …
Paper tackles medical image diagnosis with unsupervised domain adaptation.
problem Limited labeled samples and label noise in medical images.
method Collaborative Unsupervised Domain Adaptation (UDA) algorithm.
result Empirical results show superiority of the proposed method.
System suggests clinical concepts in real-time for faster note creation.
problem Efficiently creating structured clinical notes with minimal keystrokes.
method Contextual autocompletion using shallow neural networks.
result Reduces keystrokes by 67% in real hospital environments.
A novel method for classification with rejection using ensemble of cost-sensitive classifiers.
problem Avoid risky misclassification in error-critical applications.
method Learning an ensemble of cost-sensitive classifiers.
result Improved classification accuracy and flexibility in loss selection.
Study proposes a clustering and logistic regression algorithm for PU classification under Non-SCAR.
problem PU classification under Non-SCAR condition when SCAR condition is unsatisfied.
method 2-means clustering followed by logistic regression.
result Efficacy of the proposed algorithm demonstrated on 11 real data sets and a synthetic set.
The paper cleans label noise in supervised classification using Bernoulli sampling.
problem Label noise degrades supervised classifier performance.
method Proposes a label noise cleaning method based on Bernoulli random sampling.
result The method separates clean and noisy observations without prior label information.
Dual-T method improves transition matrix estimation in noisy label learning.
problem Large estimation error in noisy class posterior leads to poor transition matrix estimation.
method Introducing an intermediate class to avoid direct estimation of noisy class posterior, factorizing the transition matrix into two easier-to-estimate matrices.
result The dual-T estimator leads to better classification performances.
Unified approach for multicalibration in weakly supervised learning.
problem Existing multicalibration methods require clean input-label pairs, which are unavailable in weakly supervised learning.
method Developed estimators and post-hoc correction methods for multicalibration under weak supervision.
result Unified framework for estimating and correcting multicalibration under weak supervision with finite-sample guarantees.
Paper explores how poisoning data can increase privacy risks in machine learning models.
problem Increasing privacy risks of benign training samples through data poisoning attacks.
method Proposes generic and optimization-based attacks to amplify membership exposure.
result Demonstrates substantial increase in membership inference precision with minimal model performance degradation.
A new method handles label noise by leveraging causal information.
problem Label noise degrades deep learning performance.
method Proposes a novel generative approach using a structural causal model.
result Improves classifier performance on label-noise datasets.
In label-noise learning, \textit{noise transition matrix}, denoting the probabilities that clean labels flip into noisy labels, plays a central role in building \textit{statistically consistent classifiers}. Existing theories have shown that the transition matrix can be learned by exploiting \textit{anchor points} (i.e…
Study on noise models for noisy labels in NLP.
problem Quality of noise models from noisy labels.
method Theoretical analysis and synthetic dataset creation.
result Expected error of noise models derived.
Deep neural networks (DNNs) have been shown to over-fit a dataset when being trained with noisy labels for a long enough time. To overcome this problem, we present a simple and effective method self-ensemble label filtering (SELF) to progressively filter out the wrong labels during training. Our method improves the tas…
Generative adversarial networks (GANs) are a framework that learns a generative distribution through adversarial training. Recently, their class-conditional extensions (e.g., conditional GAN (cGAN) and auxiliary classifier GAN (AC-GAN)) have attracted much attention owing to their ability to learn the disentangled repr…
In this work, we study robust deep learning against abnormal training data from the perspective of example weighting built in empirical loss functions, i.e., gradient magnitude with respect to logits, an angle that is not thoroughly studied so far. Consequently, we have two key findings: (1) Mean Absolute Error (MAE) D…
Collecting large-scale data with clean labels for supervised training of neural networks is practically challenging. Although noisy labels are usually cheap to acquire, existing methods suffer a lot from label noise. This paper targets at the challenge of robust training at high label noise regimes. The key insight to …
We study the problem of learning conditional generators from noisy labeled samples, where the labels are corrupted by random noise. A standard training of conditional GANs will not only produce samples with wrong labels, but also generate poor quality samples. We consider two scenarios, depending on whether the noise m…
New methods protect privacy while providing accurate prediction sets.
problem Privacy-preserving conformal prediction for untrusted aggregators.
method Two LDP approaches: k-ary randomized response and binary search response.
result Finite-sample coverage guarantees and robust coverage under randomization.
Early training phase affects deep neural network optimization and generalization.
problem The choice of learning rate influences generalization in deep learning models.
method Showed that SGD implicitly penalizes the trace of the Fisher Information Matrix (FIM) from the start of training, and explicitly penalizing the trace of FIM improves generalization.
result Catastrophic Fisher explosion (large trace of FIM early in training) is linked to poor generalization.
New method improves robustness of deep learning with noisy labels.
problem Robust deep learning on corrupted labels with noisy samples.
method Meta-transition adaptation through clean meta data guidance.
result More accurate estimation of noise transition matrix and classifier parameters.
Develops NPMC method for noisy labels, improving multiclass classification accuracy.
problem Asymmetric misclassification costs and label noise in multiclass classification.
method Empirical likelihood approach using exponential tilting density ratio model.
result Root n consistent and asymptotically normal estimators for clean labels and noise mechanism.
In this work, we demonstrate universal multi-party poisoning attacks that adapt and apply to any multi-party learning process with arbitrary interaction pattern between the parties. More generally, we introduce and study (k,p)-poisoning attacks in which an adversary controls k∈[m] of the parties, and for each cor…
A new model for sequential prediction handles adversarial examples by allowing abstention.
problem Sequential prediction algorithms fail with adversarial examples, leading to incorrect predictions.
method Proposes a new model that allows abstention from predictions on adversarial examples, scaling error with VC dimension.
result A learner's error scales with the VC dimension of the hypothesis class, matching the stochastic setting.
Proposes DeGLIF to denoise graph data for label noise robustness.
problem Label noise in graph data makes node classification challenging.
method Uses leave-one-out influence function to denoise graph data.
result DeGLIF improves accuracy in node classification on noisy datasets.
Paper proposes training deep nets on noisy labels without manual annotation.
problem Training deep neural networks on noisy labels without manual annotation.
method Directly train deep neural network on noisy candidate labels, early stopping to avoid overfitting.
result Training on noisy candidate labels yields higher test performance than on clean data.
Data taggants verify dataset ownership without harming models or requiring model internals.
problem Verifying dataset ownership without harming models or requiring model internals.
method Using pairs of out-of-distribution samples and random labels as secret keys, subtly altering a dataset to detect models trained on it.
result Data taggants reliably detect models trained on the protected dataset with high confidence, without compromising validation accuracy.
Study shows label noise impacts neural representations' information content, revealing double descent behavior.
problem Impact of label noise on neural network hidden representations.
method Information Imbalance proxy of conditional mutual information to compare hidden representations.
result Representations learned with noisy labels are more informative than those with clean labels in the underparameterized regime, and equally informative in the overparameterized regime.
This work analyzes how preconditioning affects generalization in machine learning models.
problem The impact of preconditioning on the generalization of machine learning models.
method An asymptotic bias-variance decomposition of the generalization error for ridgeless regression under various preconditioners.
result The optimal preconditioner depends on label noise, model specification, and signal alignment, with NGD potentially better under certain conditions.
New methods lift weak supervision to structured prediction, providing robustness guarantees.
problem Applying weak supervision techniques to structured prediction problems.
method Introducing pseudo-Euclidean embeddings, tensor decompositions, and invariants for consistent noise rate estimation.
result Generalization guarantees nearly identical to those for models trained on clean data.
QActor optimizes learning from noisy labeled data streams by querying experts for clean labels.
problem Learning from noisy labeled data in continuous streams with limited oracle queries.
method Combines quality models for filtering and oracle queries for true labels, dynamically adjusting query limits.
result QActor nearly matches optimal accuracy with up to 6% additional ground truth data from experts.
Bayesian method improves segmentation accuracy with noisy labels.
problem Annotation errors in semantic segmentation due to mislabeling and spatial correlations.
method Approximate Bayesian estimation with spatially correlated discrete distributions and variational inference.
result The method achieves performance comparable to clean labels under moderate noise levels.