BiasICL: In-Context Learning and Demographic Biases of Vision 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_ | 1866910857607249920 |
|---|---|
| author | Xu, Sonnet Janizek, Joseph Jiang, Yixing Daneshjou, Roxana |
| author_facet | Xu, Sonnet Janizek, Joseph Jiang, Yixing Daneshjou, Roxana |
| contents | Vision language models (VLMs) show promise in medical diagnosis, but their performance across demographic subgroups when using in-context learning (ICL) remains poorly understood. We examine how the demographic composition of demonstration examples affects VLM performance in two medical imaging tasks: skin lesion malignancy prediction and pneumothorax detection from chest radiographs. Our analysis reveals that ICL influences model predictions through multiple mechanisms: (1) ICL allows VLMs to learn subgroup-specific disease base rates from prompts and (2) ICL leads VLMs to make predictions that perform differently across demographic groups, even after controlling for subgroup-specific disease base rates. Our empirical results inform best-practices for prompting current VLMs (specifically examining demographic subgroup performance, and matching base rates of labels to target distribution at a bulk level and within subgroups), while also suggesting next steps for improving our theoretical understanding of these models. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2503_02334 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | BiasICL: In-Context Learning and Demographic Biases of Vision Language Models Xu, Sonnet Janizek, Joseph Jiang, Yixing Daneshjou, Roxana Computer Vision and Pattern Recognition Artificial Intelligence Vision language models (VLMs) show promise in medical diagnosis, but their performance across demographic subgroups when using in-context learning (ICL) remains poorly understood. We examine how the demographic composition of demonstration examples affects VLM performance in two medical imaging tasks: skin lesion malignancy prediction and pneumothorax detection from chest radiographs. Our analysis reveals that ICL influences model predictions through multiple mechanisms: (1) ICL allows VLMs to learn subgroup-specific disease base rates from prompts and (2) ICL leads VLMs to make predictions that perform differently across demographic groups, even after controlling for subgroup-specific disease base rates. Our empirical results inform best-practices for prompting current VLMs (specifically examining demographic subgroup performance, and matching base rates of labels to target distribution at a bulk level and within subgroups), while also suggesting next steps for improving our theoretical understanding of these models. |
| title | BiasICL: In-Context Learning and Demographic Biases of Vision Language Models |
| topic | Computer Vision and Pattern Recognition Artificial Intelligence |
| url | https://arxiv.org/abs/2503.02334 |