[2605.11134] Spurious Correlation Learning in Preference Optimization: Mechanisms, Consequences, and Mitigation via Tie Training
Abstract:Preference learning methods like Direct Preference Optimization (DPO) are known to induce reliance on spurious correlations, leading to sycophancy and length bias in today's language models and potentially severe goal misgeneralization in future systems. In this work, we provide a unified theoretical analysis of this phenomenon, characterizing the mechanisms of spurious learning, its consequences on deployment, and a provable mitigation strategy. Focusing on log-linear policies, we show that standard preference-learning objectives induce reliance on spurious features at the population level through two channels: mean spurious bias and causal-spurious correlation leakage. We then show that this reliance creates an irreducible vulnerability to distribution shift: more data from the same training distribution fails to reduce the model's dependence on spurious features. To address this, we propose tie training, a data augmentation strategy using ties (equal-utility preference pairs) to introduce data-driven regularization. We demonstrate that this approach selectively reduces spurious learning without degrading causal learning. Finally, we validate our theory on log-linear models and provide empirical evidence that both the spurious learning mechanisms and the benefits of tie training persist for neural networks and large language models.
# link_14xcovv7y9d.pdf ## Metadata - PDFFormatVersion=1.7 - IsLinearized=false - IsAcroFormPresent=false - IsXFAPresent=false - IsCollectionPresent=false - IsSignaturesPresent=false - Author=Christian Moya; Alex Semendinger; Guang Lin; Elliott Thornley - Creator=arXiv GenPDF (tex2pdf:a6404ea) - Custom.DOI=https://doi.org/10.48550/arXiv.2605.11134 - Custom.License=http://creativecommons.org/licenses/by-nc-nd/4.0/ - Custom.PTEX.Fullbanner=This is pdfTeX, Version 3.141592653-2.6-1.40.28 (TeX Live 2025) kpathsea version 6.4.1 - Custom.arXivID=https://arxiv.org/abs/2605.11134v2 - Producer=pikepdf 8
saved by
related reading
- [2305.18290] Direct Preference Optimization: Your Language Model is Secretly a Reward Modelarxiv.org
- [2507.06187] The Delta Learning Hypothesis: Preference Tuning on Weak Data can Yield Strong Gainsarxiv.org
- RLHF & Post-Training Course by Nathan Lambertrlhfbook.com
- [2507.06187] The Delta Learning Hypothesis: Preference Tuning on Weak Data can Yield Strong Gainsarxiv.org
- Subliminal Learning: Language Models Transmit Behavioral Traits via Hidden Signals in Dataalignment.anthropic.com
- State of RL for reasoning LLMs | A. Weersaweers.de
- Understanding the Performance Gap in Preference Learning: A Dichotomy of RLHF and DPOarxiv.org
- Fine-tune Llama 2 with DPOhuggingface.co
- RLHF | John Lambertjohnwlambert.github.io
- Direct Nash Optimization: Teaching Language Models to Self-Improve with General Preferencesarxiv.org
- A Mechanistic Understanding of Alignment Algorithms: A Case Study on DPO and Toxicityarxiv.org
- Preference Tuning LLMs with Direct Preference Optimization Methodshuggingface.co