The Margin Knows the Background: Uneven Abstention in Frozen Vision-Language Models
Paper · https://arxiv.org/abs/2611.00026
arXiv:2611.00026v1 Announce Type: new Abstract: Selective prediction allows a model to abstain on uncertain inputs, but a single confidence threshold can disproportionately reject examples from particular groups. We study how confidence scores produce this imbalance in frozen vision-language models (OpenAI CLIP, LAION CLIP, SigLIP). We express a family of scores, including the margin and maximum logit, through the average and the difference of the two highest class logits, and ask which carries group information. We show that equal group coverage at every threshold is equivalent to independence between the score and group membership. A randomized margin baseline tests whether a reduction in coverage disparity beats what adding noise achieves. With score choices fixed before testing, the effect of choice depends on the dataset. On Waterbirds, the margin creates coverage gaps 2.5 to 5.5 times larger than maximum logit at 80% coverage, its group signal is in the difference component, and paired background swaps show the background moves it, although maximum logit increases overall error. On NICO++, maximum logit instead increases worst-context error by 5.4 to 6.6 points and offers no improvement over the randomized baseline for OpenAI CLIP. Across both datasets, how far a score pushes a group down its ranking tracks the coverage gap (Spearman 0.88, 0.92) better than plain group identifiability (0.61, 0.65). These results explain why a score that abstains more evenly on one dataset may be less effective on another and motivate judging group coverage and selective risk together against randomized baselines.
First seen by Oriel on .
AI summary
The study finds that confidence-score choice can sharply change group coverage disparities in frozen vision-language models, with effects varying by dataset and requiring evaluation against randomized abstention baselines.
- The analysis covers OpenAI CLIP, LAION CLIP, and SigLIP, expressing scores such as margin and maximum logit using the two highest class logits.
- On Waterbirds, margin produces coverage gaps 2.5 to 5.5 times larger than maximum logit at 80% coverage; background swaps show the background affects the margin.
- On NICO++, maximum logit raises worst-context error by 5.4 to 6.6 points and does not beat the randomized baseline for OpenAI CLIP.
- Across both datasets, how far a score pushes a group down its ranking correlates with coverage gaps more strongly than group identifiability does.
- The authors recommend assessing group coverage and selective risk together, using randomized baselines to test whether score choices help.
#vision-language models #selective prediction #group fairness #computer vision
Generated by gpt-5.6-sol. Check the original for details.
Aliases
| Source | External ID | URL |
|---|---|---|
| arXiv AI Papers | oai:arXiv.org:2611.00026v1 | https://arxiv.org/abs/2611.00026 Read discussion |