Noise Tolerance under Risk Minimization

Abstract

In this paper we explore noise tolerant learning of classifiers. We formulate the problem as follows. We assume that there is an $\mathbf{unobservable}$ training set which is noise-free. The actual training set given to the learning algorithm is obtained from this ideal data set by corrupting the class label of each example. The probability that the class label of an example is corrupted is a function of the feature vector of the example. This would account for most kinds of noisy data one encounters in practice. We say that a learning method is noise tolerant if the classifiers learnt with the ideal noise-free data and with noisy data, both have the same classification accuracy on the noise-free data. In this paper we analyze the noise tolerance properties of risk minimization (under different loss functions), which is a generic method for learning classifiers. We show that risk minimization under 0-1 loss function has impressive noise tolerance properties and that under squared error loss is tolerant only to uniform noise; risk minimization under other loss functions is not noise tolerant. We conclude the paper with some discussion on implications of these theoretical results.

Cited by in corpus (69)

• Generalized Cross Entropy Loss for Training Deep Neural Networks with Noisy Labels
• Classification with Noisy Labels by Importance Reweighting
• Deep Learning is Robust to Massive Label Noise
• Image Classification with Deep Learning in the Presence of Noisy Labels: A Survey
• Making Risk Minimization Tolerant to Label Noise
• A Light CNN for Deep Face Representation with Noisy Labels
• On the Existence of Simpler Machine Learning Models
• Double or Nothing: Multiplicative Incentive Mechanisms for Crowdsourcing
• Peer Loss Functions: Learning from Noisy Labels without Knowing Noise Rates
• On Symmetric Losses for Learning from Corrupted Labels
• Learning From Noisy Large-Scale Datasets With Minimal Supervision
• Double Ramp Loss Based Reject Option Classifier
• Machine learning \& artificial intelligence in the quantum domain
• Confidence Scores Make Instance-dependent Label-noise Learning Possible
• Fair Classification with Group-Dependent Label Noise
• Semi-Supervised Cross-Modal Retrieval with Label Prediction
• L_DMI: An Information-theoretic Noise-robust Loss Function
• Learning with Feature-Dependent Label Noise: A Progressive Approach
• Quantum-enhanced barcode decoding and pattern recognition
• Learning with Instance-Dependent Label Noise: A Sample Sieve Approach
• Robust Classification with Adiabatic Quantum Optimization
• Learning from Binary Labels with Instance-Dependent Corruption
• Mining User Behaviour from Smartphone data: a literature review
• Learning Adaptive Loss for Robust Learning with Noisy Labels
• A Second-Order Approach to Learning with Instance-Dependent Label Noise
• Webly Supervised Image Classification with Metadata: Automatic Noisy Label Correction via Visual-Semantic Graph
• Robustness of Accuracy Metric and its Inspirations in Learning with Noisy Labels
• Identifying noisy labels with a transductive semi-supervised leave-one-out filter
• Analysis of label noise in graph-based semi-supervised learning
• A Semi-Supervised Two-Stage Approach to Learning from Noisy Labels
• On the Robustness of Decision Tree Learning under Label Noise
• The Fisher-Rao Loss for Learning under Label Noise
• For self-supervised learning, Rationality implies generalization, provably
• Construction of non-convex polynomial loss functions for training a binary classifier with quantum annealing
• "Is not the truth the truth?": Analyzing the Impact of User Validations for Bus In/Out Detection in Smartphone-based Surveys
• When Optimizing $f$-divergence is Robust with Label Noise
• An Effective Label Noise Model for DNN Text Classification
• Attend in groups: a weakly-supervised deep learning framework for learning from web data
• Principled Evaluation with Human Labels: One Rater at a Time and Rater Equivalence
• Robust Deep Learning with Active Noise Cancellation for Spatial Computing
• Robust Deep Ordinal Regression Under Label Noise
• Negative-ResNet: Noisy Ambulatory Electrocardiogram Signal Classification Scheme
• Becoming More Robust to Label Noise with Classifier Diversity
• Defending against substitute model black box adversarial attacks with the 01 loss
• Defending Distributed Classifiers Against Data Poisoning Attacks
• A Computationally Efficient Classification Algorithm in Posterior Drift Model: Phase Transition and Minimax Adaptivity
• GMM Discriminant Analysis with Noisy Label for Each Class
• Product Image Recognition with Guidance Learning and Noisy Supervision
• Cost Sensitive Learning in the Presence of Symmetric Label Noise
• CCMN: A General Framework for Learning with Class-Conditional Multi-Label Noise
• Rethinking Noisy Label Models: Labeler-Dependent Noise with Adversarial Awareness
• On the transferability of adversarial examples between convex and 01 loss models
• A Survey on Deep Learning with Noisy Labels: How to train your model when you cannot trust on the annotations?
• Robust binary classification with the 01 loss
• Adversarial Poisoning Attacks and Defense for General Multi-Class Models Based On Synthetic Reduced Nearest Neighbors
• On the Robustness of Average Losses for Partial-Label Learning
• An Exploration into why Output Regularization Mitigates Label Noise
• Supervised Convex Clustering
• Temporal Calibrated Regularization for Robust Noisy Label Learning
• Towards adversarial robustness with 01 loss neural networks
• Robust Learning under Strong Noise via SQs
• Robust Trust Region for Weakly Supervised Segmentation
• A Symmetric Loss Perspective of Reliable Machine Learning
• Learning a Model for Inferring a Spatial Road Lane Network Graph using Self-Supervision
• A Non-Intrusive Correction Algorithm for Classification Problems with Corrupted Data
• Making Convex Loss Functions Robust to Outliers using $e$-Exponentiated Transformation
• Bias-Tolerant Fair Classification
• Memorization in Deep Neural Networks: Does the Loss Function matter?
• Suppressing Mislabeled Data via Grouping and Self-Attention