Spurious Correlations in Machine Learning: A Survey

2024-02-20

arXiv.org

AI Summary

Modern machine learning models suffer from spurious correlations—non-causal associations between superficial input features (e.g., background, texture) and labels—leading to poor out-of-distribution generalization and reduced robustness. To address this, we present the first systematic survey and unified methodological taxonomy for spurious correlation mitigation, integrating cross-paradigm techniques including causal inference, invariant risk minimization, environment adversarial training, debiasing regularization, and data reweighting. We rigorously define problem boundaries and establish a standardized evaluation framework. Our structured knowledge graph encompasses 120+ methods, 20+ benchmark datasets, and seven core evaluation metrics. This work delivers the first comprehensive research guide and methodological infrastructure for robust machine learning, enabling principled comparison, reproducible assessment, and informed algorithm selection in spurious correlation mitigation.

Technology Category

• Machine Learning: Adversarial Learning & Robustness
• Computer Vision: Adversarial Attacks & Robustness
• Natural Language Processing: Safety and Robustness

Application Category

• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Machine learning systems are known to be sensitive to spurious correlations between non-essential features of the inputs (e.g., background, texture, and secondary objects) and the corresponding labels. These features and their correlations with the labels are known as"spurious"because they tend to change with shifts in real-world data distributions, which can negatively impact the model's generalization and robustness. In this paper, we provide a review of this issue, along with a taxonomy of current state-of-the-art methods for addressing spurious correlations in machine learning models. Additionally, we summarize existing datasets, benchmarks, and metrics to aid future research. The paper concludes with a discussion of the recent advancements and future challenges in this field, aiming to provide valuable insights for researchers in the related domains.

Problem

• Surveying spurious correlations in machine learning models
• Addressing model sensitivity to non-essential input features
• Improving generalization and robustness against distribution shifts

Innovation

• Surveying spurious correlation mitigation techniques in ML
• Providing fine-grained taxonomy of state-of-the-art methods
• Summarizing datasets and benchmarks for future research