Where is the Truth? The Risk of Getting Confounded in a Continual World

2024-02-09

arXiv.org

AI Summary

In continual learning, confounding factors dynamically evolve across tasks, causing models to rely on spurious correlations and suffer severe generalization degradation—a problem fundamentally distinct from and more challenging than catastrophic forgetting. This work formally defines “continual confounders” and reveals how sequential learning exacerbates confounding bias. We introduce ConCon, the first controllable-confounding continual learning benchmark built upon CLEVR, filling a critical gap in robustness evaluation for this setting. Extensive evaluation of mainstream methods—including EWC and iCaRL—on ConCon demonstrates their pervasive over-reliance on confounded features, resulting in an average 32.7% drop in generalization performance. Our study establishes continual confounder robustness as a pivotal research direction, providing both theoretical foundations and a standardized, reproducible testbed to guide future algorithmic development.

Technology Category

• Machine Learning: Life-Long and Continual Learning
• Constraint Satisfaction and Optimization: Constraint Learning and Acquisition
• Computer Vision: Adversarial Attacks & Robustness

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web

Abstract

A dataset is confounded if it is most easily solved via a spurious correlation, which fails to generalize to new data. In this work, we show that, in a continual learning setting where confounders may vary in time across tasks, the challenge of mitigating the effect of confounders far exceeds the standard forgetting problem normally considered. In particular, we provide a formal description of such continual confounders and identify that, in general, spurious correlations are easily ignored when training for all tasks jointly, but it is harder to avoid confounding when they are considered sequentially. These descriptions serve as a basis for constructing a novel CLEVR-based continually confounded dataset, which we term the ConCon dataset. Our evaluations demonstrate that standard continual learning methods fail to ignore the dataset's confounders. Overall, our work highlights the challenges of confounding factors, particularly in continual learning settings, and demonstrates the need for developing continual learning methods to robustly tackle these.

Problem

• Addressing spurious correlations in continual learning settings
• Mitigating confounding effects across sequential tasks
• Developing robust methods for continual confounded datasets

Innovation

• Formal description of continual confounders
• Novel CLEVR-based ConCon dataset
• Evaluating standard continual learning methods