Striking the Right Balance with Uncertainty

Abstract

Learning unbiased models on imbalanced datasets is a significant challenge. Rare classes tend to get a concentrated representation in the classification space which hampers the generalization of learned boundaries to new test examples. In this paper, we demonstrate that the Bayesian uncertainty estimates directly correlate with the rarity of classes and the difficulty level of individual samples. Subsequently, we present a novel framework for uncertainty based class imbalance learning that follows two key insights: First, classification boundaries should be extended further away from a more uncertain (rare) class to avoid overfitting and enhance its generalization. Second, each sample should be modeled as a multi-variate Gaussian distribution with a mean vector and a covariance matrix defined by the sample's uncertainty. The learned boundaries should respect not only the individual samples but also their distribution in the feature space. Our proposed approach efficiently utilizes sample and class uncertainty information to learn robust features and more generalizable classifiers. We systematically study the class imbalance problem and derive a novel loss formulation for max-margin learning based on Bayesian uncertainty measure. The proposed method shows significant performance improvements on six benchmark datasets for face verification, attribute prediction, digit/object classification and skin lesion detection.

arXiv commentCVPR 2019

References in corpus (11)

• SMOTE: Synthetic Minority Over-sampling Technique
• Deep Learning Face Representation by Joint Identification-Verification
• Multi-Task Learning Using Uncertainty to Weigh Losses for Scene Geometry and Semantics
• SphereFace: Deep Hypersphere Embedding for Face Recognition
• Detecting Adversarial Samples from Artifacts
• Targeting Ultimate Accuracy: Face Recognition via Deep Embedding
• Batch-normalized Maxout Network in Network
• Do We Really Need to Collect Millions of Faces for Effective Face Recognition?
• Linear Maximum Margin Classifier for Learning from Uncertain Data
• Feature Transfer Learning for Deep Face Recognition with Under-Represented Data
• Max-margin Class Imbalanced Learning with Gaussian Affinity

Cited by in corpus (14)

• A Review of Uncertainty Quantification in Deep Learning: Techniques, Applications and Challenges
• iSAID: A Large-scale Dataset for Instance Segmentation in Aerial Images
• Deep Long-Tailed Learning: A Survey
• Distribution-Balanced Loss for Multi-Label Classification in Long-Tailed Datasets
• Distribution-aware Margin Calibration for Semantic Segmentation in Images
• Jointly De-biasing Face Recognition and Demographic Attribute Estimation
• Point Cloud Instance Segmentation using Probabilistic Embeddings
• Multi-label Contrastive Predictive Coding
• Distributional Robustness Loss for Long-tail Learning
• F^2-Softmax: Diversifying Neural Text Generation via Frequency Factorized Softmax
• Distribution-aware Margin Calibration for Medical Image Segmentation
• Calibrating Class Activation Maps for Long-Tailed Visual Recognition
• On-target Adaptation
• Imbalance Robust Softmax for Deep Embeeding Learning