No Subclass Left Behind: Fine-Grained Robustness in Coarse-Grained Classification Problems

Abstract

In real-world classification tasks, each class often comprises multiple finer-grained "subclasses." As the subclass labels are frequently unavailable, models trained using only the coarser-grained class labels often exhibit highly variable performance across different subclasses. This phenomenon, known as hidden stratification, has important consequences for models deployed in safety-critical applications such as medicine. We propose GEORGE, a method to both measure and mitigate hidden stratification even when subclass labels are unknown. We first observe that unlabeled subclasses are often separable in the feature space of deep neural networks, and exploit this fact to estimate subclass labels for the training data via clustering techniques. We then use these approximate subclass labels as a form of noisy supervision in a distributionally robust optimization objective. We theoretically characterize the performance of GEORGE in terms of the worst-case generalization error across any subclass. We empirically validate GEORGE on a mix of real-world and benchmark image classification datasets, and show that our approach boosts worst-case subclass accuracy by up to 22 percentage points compared to standard training techniques, without requiring any prior information about the subclasses.

40 pages. Published as a conference paper at NeurIPS 2020

Cited by in corpus (16)

• WILDS: A Benchmark of in-the-Wild Distribution Shifts
• Where Does My Model Underperform? A Human Evaluation of Slice Discovery Algorithms
• Environment Inference for Invariant Learning
• Simple data balancing achieves competitive worst-group-accuracy
• Camera-aware Proxies for Unsupervised Person Re-Identification
• Representation Matters: Assessing the Importance of Subgroup Allocations in Training Data
• Model-Based Domain Generalization
• Learning Debiased and Disentangled Representations for Semantic Segmentation
• Robustness in Compressed Neural Networks for Object Detection
• Teacher's pet: understanding and mitigating biases in distillation
• Can Querying for Bias Leak Protected Attributes? Achieving Privacy With Smooth Sensitivity
• Unsupervised Learning of Debiased Representations with Pseudo-Attributes
• Improving Group Robustness on Spurious Correlation via Evidential Alignment
• Training Over-parameterized Models with Non-decomposable Objectives
• A Topological-Framework to Improve Analysis of Machine Learning Model Performance
• Combating Unknown Bias with Effective Bias-Conflicting Scoring and Gradient Alignment