Classification with Noisy Labels by Importance Reweighting

Abstract

In this paper, we study a classification problem in which sample labels are randomly corrupted. In this scenario, there is an unobservable sample with noise-free labels. However, before being observed, the true labels are independently flipped with a probability $\rho \in [0,0.5)$, and the random label noise can be class-conditional. Here, we address two fundamental problems raised by this scenario. The first is how to best use the abundant surrogate loss functions designed for the traditional classification problem when there is label noise. We prove that any surrogate loss function can be used for classification with noisy labels by using importance reweighting, with consistency assurance that the label noise does not ultimately hinder the search for the optimal classifier of the noise-free sample. The other is the open problem of how to obtain the noise rate $\rho$. We show that the rate is upper bounded by the conditional probability $P(y|x)$ of the noisy sample. Consequently, the rate can be estimated, because the upper bound can be easily reached in classification problems. Experimental results on synthetic and real datasets confirm the efficiency of our methods.

References in corpus (2)

• Multiple Kernel Learning from Noisy Labels by Stochastic Programming
• Constructive Setting of the Density Ratio Estimation Problem and its Rigorous Solution

Cited by in corpus (139)

• Learning from positive and unlabeled data: a survey
• Robust Face Recognition via Multimodal Deep Face Representation
• Multi-Directional Multi-Level Dual-Cross Patterns for Robust Face Recognition
• Image Classification with Deep Learning in the Presence of Noisy Labels: A Survey
• Delving Deep into Label Smoothing
• Hyperspectral Image Classification in the Presence of Noisy Labels
• Are Anchor Points Really Indispensable in Label-Noise Learning?
• Pervasive AI for IoT applications: A Survey on Resource-efficient Distributed Artificial Intelligence
• How does Disagreement Help Generalization against Label Corruption?
• Making Deep Neural Networks Robust to Label Noise: a Loss Correction Approach
• Unsupervised Domain Adaptation with Adversarial Residual Transform Networks
• Large Margin Multi-modal Multi-task Feature Extraction for Image Classification
• Dual T: Reducing Estimation Error for Transition Matrix in Label-noise Learning
• A Survey of Label-noise Representation Learning: Past, Present and Future
• Unsupervised Embedding Learning via Invariant and Spreading Instance Feature
• Self-Guided Learning to Denoise for Robust Recommendation
• Combating noisy labels by agreement: A joint training method with co-regularization
• Deep Learning Driven Visual Path Prediction from a Single Image
• Learning with Noisy Labels Revisited: A Study Using Real-World Human Annotations
• Peer Loss Functions: Learning from Noisy Labels without Knowing Noise Rates
• Learning from Noisy Labels with Distillation
• Joint Hierarchical Category Structure Learning and Large-Scale Image Classification
• Co-Correcting: Noise-tolerant Medical Image Classification via mutual Label Correction
• Unsupervised Learning with Stein's Unbiased Risk Estimator
• Compete to Win: Enhancing Pseudo Labels for Barely-supervised Medical Image Segmentation
• On the Minimal Supervision for Training Any Binary Classifier from Only Unlabeled Data
• Linear Maximum Margin Classifier for Learning from Uncertain Data
• Noisy multi-label semi-supervised dimensionality reduction
• Loss factorization, weakly supervised learning and label noise robustness
• Confidence Scores Make Instance-dependent Label-noise Learning Possible
• Fair Classification with Group-Dependent Label Noise
• Boosting Facial Expression Recognition by A Semi-Supervised Progressive Teacher
• L_DMI: An Information-theoretic Noise-robust Loss Function
• Noise-tolerant fair classification
• Mitigating Overfitting in Supervised Classification from Two Unlabeled Datasets: A Consistent Risk Correction Approach
• Open-set Label Noise Can Improve Robustness Against Inherent Label Noise
• Instance-Dependent PU Learning by Bayesian Optimal Relabeling
• Quantum-enhanced barcode decoding and pattern recognition
• What Do Neural Networks Learn When Trained With Random Labels?
• Trustable Co-label Learning from Multiple Noisy Annotators
• Learning with Bounded Instance- and Label-dependent Label Noise
• Learning with Instance-Dependent Label Noise: A Sample Sieve Approach
• Provably End-to-end Label-Noise Learning without Anchor Points
• Robust Long-Tailed Learning under Label Noise
• Extended T: Learning with Mixed Closed-set and Open-set Noisy Labels
• Learning Noise Transition Matrix from Only Noisy Labels via Total Variation Regularization
• Fairness Evaluation in Presence of Biased Noisy Labels
• Uncertainty Based Detection and Relabeling of Noisy Image Labels
• SCE: A manifold regularized set-covering method for data partitioning
• Penalty Method for Inversion-Free Deep Bilevel Optimization
• Clusterability as an Alternative to Anchor Points When Learning with Noisy Labels
• Heteroskedastic and Imbalanced Deep Learning with Adaptive Regularization
• Certified Robustness to Label-Flipping Attacks via Randomized Smoothing
• Distributed Deep Forest and its Application to Automatic Detection of Cash-out Fraud
• Simple and Effective Regularization Methods for Training on Noisily Labeled Data with Generalization Guarantee
• Variational Rectification Inference for Learning with Noisy Labels
• DistillHash: Unsupervised Deep Hashing by Distilling Data Pairs
• Learnability for the Information Bottleneck
• Rethinking Importance Weighting for Deep Learning under Distribution Shift
• A Second-Order Approach to Learning with Instance-Dependent Label Noise
• Counterfactual Prediction Under Outcome Measurement Error
• Learning the Synthesizability of Dynamic Texture Samples
• In Defense of the Triplet Loss Again: Learning Robust Person Re-Identification with Fast Approximated Triplet Loss and Label Distillation
• Searching to Exploit Memorization Effect in Learning from Corrupted Labels
• Weakly Supervised AUC Optimization: A Unified Partial AUC Approach
• SnapMix: Semantically Proportional Mixing for Augmenting Fine-grained Data
• Learning Fast Sample Re-weighting Without Reward Data
• The Rich Get Richer: Disparate Impact of Semi-Supervised Learning
• Federated Noisy Client Learning
• Fair Classification with Noisy Protected Attributes: A Framework with Provable Guarantees
• SOSELETO: A Unified Approach to Transfer Learning and Training with Noisy Labels
• PNT-Edge: Towards Robust Edge Detection with Noisy Labels by Learning Pixel-level Noise Transitions
• Butterfly: One-step Approach towards Wildly Unsupervised Domain Adaptation
• Policy Learning Using Weak Supervision
• Learning Noise-Aware Encoder-Decoder from Noisy Labels by Alternating Back-Propagation for Saliency Detection
• Approximating Instance-Dependent Noise via Instance-Confidence Embedding
• When Optimizing $f$-divergence is Robust with Label Noise
• Non-Negative Bregman Divergence Minimization for Deep Direct Density Ratio Estimation
• Which Strategies Matter for Noisy Label Classification? Insight into Loss and Uncertainty
• Generalized Data Weighting via Class-level Gradient Manipulation
• Multi-Class Classification from Noisy-Similarity-Labeled Data
• Bayesian Statistics Guided Label Refurbishment Mechanism: Mitigating Label Noise in Medical Image Classification
• Less is More: Learning Highlight Detection from Video Duration
• Robust and On-the-fly Dataset Denoising for Image Classification
• Affective Image Content Analysis: Two Decades Review and New Perspectives
• Learning Multiclass Classifier Under Noisy Bandit Feedback
• Revisiting Sample Selection Approach to Positive-Unlabeled Learning: Turning Unlabeled Data into Positive rather than Negative
• A Novel Self-Supervised Re-labeling Approach for Training with Noisy Labels
• Noise-Resilient Point-wise Anomaly Detection in Time Series Using Weak Segment Labels
• Learning to Combat Noisy Labels via Classification Margins
• Importance Reweighting for Biquality Learning
• Can Less be More? When Increasing-to-Balancing Label Noise Rates Considered Beneficial
• Extreme Multi-label Classification from Aggregated Labels
• NGC: A Unified Framework for Learning with Open-World Noisy Data
• No Regret Sample Selection with Noisy Labels
• KNN-enhanced Deep Learning Against Noisy Labels
• Co-matching: Combating Noisy Labels by Augmentation Anchoring
• Deep k-NN for Noisy Labels
• General Supervision via Probabilistic Transformations
• Intelligence, physics and information -- the tradeoff between accuracy and simplicity in machine learning
• Robust Deep Ordinal Regression Under Label Noise
• Learning with Noisy Labels by Efficient Transition Matrix Estimation to Combat Label Miscorrection
• Adversarial-Based Knowledge Distillation for Multi-Model Ensemble and Noisy Data Refinement
• Partial Multi-label Learning with Label and Feature Collaboration
• EQFace: A Simple Explicit Quality Network for Face Recognition
• Defending Distributed Classifiers Against Data Poisoning Attacks
• Cost Sensitive Learning in the Presence of Symmetric Label Noise
• Learning from Noisy Labels with Noise Modeling Network
• Utilizing Grounded SAM for self-supervised frugal camouflaged human detection
• Constrained Instance and Class Reweighting for Robust Learning under Label Noise
• Graph convolutional networks for learning with few clean and many noisy labels
• Transform consistency for learning with noisy labels
• Binary Classification from Multiple Unlabeled Datasets via Surrogate Set Classification
• Multiplicative Reweighting for Robust Neural Network Optimization
• P-DIFF: Learning Classifier with Noisy Labels based on Probability Difference Distributions
• Leveraging Local Variation in Data: Sampling and Weighting Schemes for Supervised Deep Learning
• Combating Label Noise With A General Surrogate Model For Sample Selection
• New Metric Formulas that Include Measurement Errors in Machine Learning for Natural Sciences
• Learning with partially separable data
• Theoretical Guarantees of Deep Embedding Losses Under Label Noise
• Compensation Learning
• A Non-Intrusive Correction Algorithm for Classification Problems with Corrupted Data
• TEMImageNet Training Library and AtomSegNet Deep-Learning Models for High-Precision Atom Segmentation, Localization, Denoising, and Super-Resolution Processing of Atomic-Resolution Images
• Contrastive Representations for Label Noise Require Fine-Tuning
• The Impact of Label Noise on a Music Tagger
• Alleviating Noisy-label Effects in Image Classification via Probability Transition Matrix
• Collective Learning
• Binary Classification of Light and Dark Time Traces of a Transition Edge Sensor Using Convolutional Neural Networks
• Analysis of classifiers robust to noisy labels
• Bias-Tolerant Fair Classification
• Robust Online Convex Optimization in the Presence of Outliers
• Unaligned Image-to-Image Translation by Learning to Reweight
• Mitigating Memorization in Sample Selection for Learning with Noisy Labels
• Exploring Representativeness and Informativeness for Active Learning
• Query-Adaptive Hash Code Ranking for Large-Scale Multi-View Visual Search
• Learning with Noisy Labels for Sentence-level Sentiment Classification
• Supervised Learning in the Presence of Noise: Application in ICD-10 Code Classification
• Batch Inverse-Variance Weighting: Deep Heteroscedastic Regression
• Learning from Web Data with Self-Organizing Memory Module