# Universal Boosts, Specific Suppressors: Sparse Autoencoder Steering of Medical Vision-Language Models

arXiv: https://arxiv.org/abs/2605.24977 · HTML: https://mlsanity.com/paper/2605.24977/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Medical vision-language models, one of 16 directions in Multimodal LLMs / vision-language models.
- Builds on 15 of its 35 references that are in the map.
- First posted on arXiv 24 May 2026.

## Placed in

- Multimodal LLMs / vision-language models / [Medical vision-language models](https://mlsanity.com/mllm/medical-vision-language-models/)
- Mechanistic interpretability / [Interpreting biomedical and science models](https://mlsanity.com/mechinterp/interpreting-biomedical-and-science-models/)

## Builds on (most cited first, 15 of 15)

- [BERTScore: Evaluating Text Generation with BERT](https://mlsanity.com/paper/1904.09675/) (2019)
- [Sparse Autoencoders Find Highly Interpretable Features in Language Models](https://mlsanity.com/paper/2309.08600/) (2023)
- [Inference-Time Intervention: Eliciting Truthful Answers from a Language Model](https://mlsanity.com/paper/2306.03341/) (2023)
- [Mitigating Object Hallucinations in Large Vision-Language Models through Visual Contrastive Decoding](https://mlsanity.com/paper/2311.16922/) (2023)
- [OPERA: Alleviating Hallucination in Multi-Modal Large Language Models via Over-Trust Penalty and Retrospection-Allocation](https://mlsanity.com/paper/2311.17911/) (2023)
- [Sparse Feature Circuits: Discovering and Editing Interpretable Causal Graphs in Language Models](https://mlsanity.com/paper/2403.19647/) (2024)
- [CheXbert: Combining Automatic Labelers and Expert Annotations for Accurate Radiology Report Labeling Using BERT](https://mlsanity.com/paper/2004.09167/) (2020)
- [RadGraph: Extracting Clinical Entities and Relations from Radiology Reports](https://mlsanity.com/paper/2106.14463/) (2021)
- [GREEN: Generative Radiology Report Evaluation and Error Notation](https://mlsanity.com/paper/2405.03595/) (2024)
- [Steering Language Model Refusal with Sparse Autoencoders](https://mlsanity.com/paper/2411.11296/) (2024)
- [SAEs Are Good for Steering -- If You Select the Right Features](https://mlsanity.com/paper/2505.20063/) (2025)
- [Steering Large Language Models using Conceptors: Improving Addition-Based Activation Engineering](https://mlsanity.com/paper/2410.16314/) (2024)
- [RadVLM: A Multitask Conversational Vision-Language Model for Radiology](https://mlsanity.com/paper/2502.03333/) (2025)
- [RadEval: A framework for radiology text evaluation](https://mlsanity.com/paper/2509.18030/) (2025)
- [CheXalign: Preference fine-tuning in chest X-ray interpretation models without human feedback](https://mlsanity.com/paper/2410.07025/) (2024)

## Abstract (arXiv)

Medical vision-language models (VLMs) often hallucinate findings when generating chest X-ray reports: they fabricate findings that are not present in the image, miss important ones, or locate them incorrectly. We mitigate this without weight updates by decoding-time residual steering on a per-token sparse autoencoder (SAE) basis: Top-K SAEs on late layers, causal steering against clinical errors, then combined suppress/boost intervention at inference time. On the MIMIC-CXR test split, our inference-only method improves the quality of generated reports for three radiology VLMs (RadVLM, LLaVA-Rad, and CheXOne), with relative improvements of +5.4%, +7.2%, and +17.0% in the clinical composite metric, and statistically significant GREEN gains on all backbones. A cross-model feature alignment shows that the quality-promoting (boost) directions overlap strongly across architectures, whereas hallucination-linked (suppress) directions are model-specific. Therefore, transferable steering must treat suppression per-backbone, rather than sharing a universal suppress list. The same recipe transfers zero-shot to IU-Xray (Green +7.7% rel.) without retraining, confirming that the identified features are properties of the model, not of the training corpus. We release causal feature sets and an interactive feature dashboard: https://cxr-sparse-feature-dashboard.netlify.app/.