---
title: "BiRG-LoRA: Clinically Structured Rank-Gated Low-Rank Adaptation for Cross-Benchmark Medical Question Answering"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2606.31432v1"
permalink: "https://feed.astralor.com/items/itm_8a028c0ac2454d3b7e566ecf"
published: "2026-06-30T09:59:05.000Z"
score: 75
channel: "ai"
tags: ["BiRG-LoRA","Qwen3","Low-Rank Adaptation","Medical QA","Gating Mechanism"]
---

# BiRG-LoRA: Clinically Structured Rank-Gated Low-Rank Adaptation for Cross-Benchmark Medical Question Answering

## AI commentary

> This paper is particularly worth reading for researchers in medical NLP and parameter-efficient fine-tuning: it directly integrates clinical priors into LoRA's rank selection, not as a mixture-of-experts but by dynamically adjusting a single adapter's expressiveness per input—a clever design. The experiments are rigorous with proper statistical testing, though only single-seed. For practitioners seeking low-cost adaptation of large models in medical products, this parameter-reducing yet effective solution offers a valuable reference.

## At a glance

This paper proposes BiRG-LoRA, a single-adapter rank-gated LoRA method tailored for medical multiple-choice question answering. It employs a biaxial gate mechanism that combines hidden semantic evidence with clinical specialty/operation priors to dynamically select sparse top-k rank atoms, with a scalar injection coefficient controlling adapter strength. Evaluated on Qwen3-8B, BiRG-LoRA achieves 69.31% macro-average accuracy across CMB, CMExam, MedQA, and MedMCQA, outperforming MoELoRA by 0.89 percentage points while using 28.1% fewer trainable parameters. The results demonstrate that clinically structured rank allocation effectively improves cross-benchmark medical QA.

## Key points

- Proposes BiRG-LoRA, a single-adapter rank-gated LoRA that conditions on input via biaxial gate (hidden semantics + clinical priors) to select sparse top-k rank atoms and control adapter strength.
- Achieves 69.31% macro-average accuracy across four medical QA benchmarks using Qwen3-8B, surpassing MoELoRA by 0.89 pp with 28.1% fewer parameters.
- Outperforms vanilla LoRA r16 and rank-matched LoRA r4 by 0.83 macro points, and remains robust to moderate tag noise.

## Full text

Medical multiple-choice question answering requires parameter-efficient adaptation across heterogeneous knowledge domains and reasoning operations. A medication question, a diagnostic decision, a public-health item, and a nursing-action item may require different low-rank updates, while some recall items should preserve the base model's representation with only mild adapter intervention. We propose BiRG-LoRA, a single-adapter rank-gated LoRA method for medical question answering. BiRG-LoRA keeps one LoRA module per target layer but makes its rank dimension input-conditioned: for each question, a biaxial gate combines hidden semantic evidence with specialty/profession priors, clinical-operation priors, and their interaction to select a sparse top-$k$ subset of rank atoms. A scalar injection coefficient further controls the strength of the selected adapter update. Under a matched Qwen3-8B CMB-source protocol, BiRG-LoRA achieves the highest four-benchmark macro-average accuracy among trainable PEFT baselines and matched routing controls: 69.31% averaged over CMB, CMExam, MedQA, and MedMCQA. It improves over MoELoRA by 0.89 percentage points while using 28.1% fewer trainable parameters; a paired, benchmark-stratified bootstrap over final predictions gives a 95% confidence interval of [0.42, 1.37] for this macro-average gain. Basic controls show that BiRG-LoRA also improves over vanilla LoRA r16 and active-rank-matched LoRA r4 by 0.83 macro points, and an evaluation-time weak-axis perturbation check suggests that performance is not brittle to moderate tag noise. The results support a bounded claim: clinically structured rank allocation improves cross-benchmark medical QA under a matched single-seed protocol, while training-seed variance remains future work.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_8a028c0ac2454d3b7e566ecf