ACE and Diverse Generalization via Selective Disagreement

2025-09-09

AI Summary

Deep neural networks suffer from spurious correlations, leading to fundamental generalization uncertainty—especially under fully spurious correlation settings. To address this, we propose a robust concept learning framework based on selective disagreement: it introduces high-confidence constraints and a selective disagreement mechanism to leverage prediction inconsistencies over unlabeled data during self-training, enabling unsupervised model selection and integration of prior knowledge. Crucially, our method mitigates underspecification without requiring trusted labels or unreliable external metrics. Evaluated on both complete and incomplete spurious correlation benchmarks, it achieves state-of-the-art performance. Moreover, it demonstrates strong generalization and configuration flexibility in language model alignment tasks, confirming its broad applicability beyond standard vision-centric settings.

Technology Category

• Machine Learning: Adversarial Learning & Robustness
• Natural Language Processing: Safety and Robustness
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Deep neural networks are notoriously sensitive to spurious correlations - where a model learns a shortcut that fails out-of-distribution. Existing work on spurious correlations has often focused on incomplete correlations,leveraging access to labeled instances that break the correlation. But in cases where the spurious correlations are complete, the correct generalization is fundamentally extit{underspecified}. To resolve this underspecification, we propose learning a set of concepts that are consistent with training data but make distinct predictions on a subset of novel unlabeled inputs. Using a self-training approach that encourages extit{confident} and extit{selective} disagreement, our method ACE matches or outperforms existing methods on a suite of complete-spurious correlation benchmarks, while remaining robust to incomplete spurious correlations. ACE is also more configurable than prior approaches, allowing for straight-forward encoding of prior knowledge and principled unsupervised model selection. In an early application to language-model alignment, we find that ACE achieves competitive performance on the measurement tampering detection benchmark extit{without} access to untrusted measurements. While still subject to important limitations, ACE represents significant progress towards overcoming underspecification.

Problem

• Resolving underspecification in complete spurious correlations
• Learning concepts making distinct predictions on novel inputs
• Overcoming model sensitivity to distributional shortcuts

Innovation

• Self-training with confident selective disagreement
• Learning consistent concepts with distinct predictions
• Configurable prior knowledge encoding and model selection