# Mechanistically Guided LoRA Improves Paraphrase Consistency in Medical Vision-Language Models

arXiv: https://arxiv.org/abs/2603.00148 · HTML: https://mlsanity.com/paper/2603.00148/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- Placed in Medical vision-language models, one of 16 directions in Multimodal LLMs / vision-language models.
- Builds on 10 of its 25 references that are in the map.
- First posted on arXiv 24 Feb 2026.

## Placed in

- Multimodal LLMs / vision-language models / [Medical vision-language models](https://mlsanity.com/mllm/medical-vision-language-models/)

## Builds on (most cited first, 10 of 10)

- [Locating and Editing Factual Associations in GPT](https://mlsanity.com/paper/2202.05262/) (2022)
- [Sparse Autoencoders Find Highly Interpretable Features in Language Models](https://mlsanity.com/paper/2309.08600/) (2023)
- [LLaVA-Med: Training a Large Language-and-Vision Assistant for Biomedicine in One Day](https://mlsanity.com/paper/2306.00890/) (2023)
- [Beyond Accuracy: Behavioral Testing of NLP models with CheckList](https://mlsanity.com/paper/2005.04118/) (2020)
- [Publicly Available Clinical BERT Embeddings](https://mlsanity.com/paper/1904.03323/) (2019)
- [SLAKE: A Semantically-Labeled Knowledge-Enhanced Dataset for Medical Visual Question Answering](https://mlsanity.com/paper/2102.09542/) (2021)
- [Sparse Feature Circuits: Discovering and Editing Interpretable Causal Graphs in Language Models](https://mlsanity.com/paper/2403.19647/) (2024)
- [Measuring and Improving Consistency in Pretrained Language Models](https://mlsanity.com/paper/2102.01017/) (2021)
- [PadChest: A large chest x-ray image dataset with multi-label annotated reports](https://mlsanity.com/paper/1901.07441/) (2019)
- [Cycle-Consistency for Robust Visual Question Answering](https://mlsanity.com/paper/1902.05660/) (2019)

## Built on by (most cited first, 2 of 2)

- [Predictive Entropy as a Joint Screen for Error and Paraphrase Instability in Medical Vision-Language Models](https://mlsanity.com/paper/2604.08941/) (2026)
- [Attention Without Grounding: Causal Evaluation of Visual Explanations in Medical VLMs](https://mlsanity.com/paper/2607.18577/) (2026)

## Abstract (arXiv)

Medical Vision-Language Models can give different yes or no answers to rephrasings of the same clinical question. We study this in MedGemma-4B using PSF-Med Sadanandan and Behzadan (2025), which provides paraphrase pairs for systematic consistency evaluation on medical VQA. On MIMIC-CXR binary questions (n = 158), the baseline flip rate is 14.6% and mean margin difference is 1.63 logits. We validate that Gemma Scope 2 Sparse Autoencoders (SAEs) transfer to MedGemma activations, achieving R2 ~= 0.997 on both medical and general text (n = 100 prompts each, p < 0.001 for exceeding a 0.95 threshold). We then fine-tune Low-Rank Adaptation (LoRA) adapters with a combined loss that balances paraphrase consistency with answer accuracy. This combined approach prevents mode collapse that occurs with pure consistency training while reducing flip rate from 14.6% to 4.4% (p = 0.002, two-proportion z-test) and margin difference from 1.63 to 0.33 (79.5% reduction). Accuracy remains stable at 84.2% baseline versus 82.3% after training (-1.9pp, not significant). On PadChest Balanced (n = 250), flip rate drops from 13.6% to 7.8%, mean margin difference drops from 1.08 to 0.35 (67.9% reduction), and accuracy increases from 66.4% to 69.4%. A layer-range ablation shows that early layers reduce margin differences more than mechanistically selected middle layers.