Assessing Robustness to Spurious Correlations in Post-Training Language Models
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866910934713237504 |
|---|---|
| author | Shuieh, Julia Singhal, Prasann Shanker, Apaar Heyer, John Pu, George Denton, Samuel |
| author_facet | Shuieh, Julia Singhal, Prasann Shanker, Apaar Heyer, John Pu, George Denton, Samuel |
| contents | Supervised and preference-based fine-tuning techniques have become popular for aligning large language models (LLMs) with user intent and correctness criteria. However, real-world training data often exhibits spurious correlations -- arising from biases, dataset artifacts, or other "shortcut" features -- that can compromise a model's performance or generalization. In this paper, we systematically evaluate three post-training algorithms -- Supervised Fine-Tuning (SFT), Direct Preference Optimization (DPO), and KTO (Kahneman-Tversky Optimization) -- across a diverse set of synthetic tasks and spuriousness conditions. Our tasks span mathematical reasoning, constrained instruction-following, and document-grounded question answering. We vary the degree of spurious correlation (10% vs. 90%) and investigate two forms of artifacts: "Feature Ambiguity" and "Distributional Narrowness." Our results show that the models often but not always degrade under higher spuriousness. The preference-based methods (DPO/KTO) can demonstrate relative robustness in mathematical reasoning tasks. By contrast, SFT maintains stronger performance in complex, context-intensive tasks. These findings highlight that no single post-training strategy universally outperforms in all scenarios; the best choice depends on the type of target task and the nature of spurious correlations. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2505_05704 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Assessing Robustness to Spurious Correlations in Post-Training Language Models Shuieh, Julia Singhal, Prasann Shanker, Apaar Heyer, John Pu, George Denton, Samuel Computation and Language Artificial Intelligence Supervised and preference-based fine-tuning techniques have become popular for aligning large language models (LLMs) with user intent and correctness criteria. However, real-world training data often exhibits spurious correlations -- arising from biases, dataset artifacts, or other "shortcut" features -- that can compromise a model's performance or generalization. In this paper, we systematically evaluate three post-training algorithms -- Supervised Fine-Tuning (SFT), Direct Preference Optimization (DPO), and KTO (Kahneman-Tversky Optimization) -- across a diverse set of synthetic tasks and spuriousness conditions. Our tasks span mathematical reasoning, constrained instruction-following, and document-grounded question answering. We vary the degree of spurious correlation (10% vs. 90%) and investigate two forms of artifacts: "Feature Ambiguity" and "Distributional Narrowness." Our results show that the models often but not always degrade under higher spuriousness. The preference-based methods (DPO/KTO) can demonstrate relative robustness in mathematical reasoning tasks. By contrast, SFT maintains stronger performance in complex, context-intensive tasks. These findings highlight that no single post-training strategy universally outperforms in all scenarios; the best choice depends on the type of target task and the nature of spurious correlations. |
| title | Assessing Robustness to Spurious Correlations in Post-Training Language Models |
| topic | Computation and Language Artificial Intelligence |
| url | https://arxiv.org/abs/2505.05704 |