@inproceedings{zhang-etal-2026-pr,
title = "{PR}-{L}o{RA}: Direction-Regularized Low-Rank Adaptation for Behavioral Alignment in Large Language Models",
author = "Zhang, Yikai and
Xu, Rui and
Chen, Aili and
Zhang, Haoyu and
Wu, Libo and
Yinghui, Xu and
Xiao, Yanghua",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.2317/",
pages = "52640--52654",
ISBN = "979-8-89176-459-0",
abstract = "Behavioral problems in deployed LLMs are often discovered when only limited supervision is available and repeating full alignment training is impractical. Existing post-hoc methods have complementary limitations: activation steering lacks input adaptivity, while standard LoRA learns without an explicit behavioral direction prior and can be unstable in low-data settings. We introduce PR-LoRA, a direction-regularized LoRA variant that initializes LoRA $B$ matrices with per-layer contrastive behavioral directions and anchors them with an $L_2$ penalty, while leaving $A$ freely trainable. We evaluate PR-LoRA on seven CAA behaviors across Qwen3 and LLaMA models against activation steering and parameter-efficient adaptation baselines. On Qwen3-8B, PR-LoRA lowers the seven-behavior macro-average undesirable-behavior score from .405 to .374 relative to standard LoRA, a 7.8{\%} relative reduction, and outperforms DPO-LoRA and LoReFT under the tested low-update budget."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zhang-etal-2026-pr">
<titleInfo>
<title>PR-LoRA: Direction-Regularized Low-Rank Adaptation for Behavioral Alignment in Large Language Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yikai</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Rui</namePart>
<namePart type="family">Xu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Aili</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Haoyu</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Libo</namePart>
<namePart type="family">Wu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xu</namePart>
<namePart type="family">Yinghui</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yanghua</namePart>
<namePart type="family">Xiao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>Behavioral problems in deployed LLMs are often discovered when only limited supervision is available and repeating full alignment training is impractical. Existing post-hoc methods have complementary limitations: activation steering lacks input adaptivity, while standard LoRA learns without an explicit behavioral direction prior and can be unstable in low-data settings. We introduce PR-LoRA, a direction-regularized LoRA variant that initializes LoRA B matrices with per-layer contrastive behavioral directions and anchors them with an L₂ penalty, while leaving A freely trainable. We evaluate PR-LoRA on seven CAA behaviors across Qwen3 and LLaMA models against activation steering and parameter-efficient adaptation baselines. On Qwen3-8B, PR-LoRA lowers the seven-behavior macro-average undesirable-behavior score from .405 to .374 relative to standard LoRA, a 7.8% relative reduction, and outperforms DPO-LoRA and LoReFT under the tested low-update budget.</abstract>
<identifier type="citekey">zhang-etal-2026-pr</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.2317/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>52640</start>
<end>52654</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T PR-LoRA: Direction-Regularized Low-Rank Adaptation for Behavioral Alignment in Large Language Models
%A Zhang, Yikai
%A Xu, Rui
%A Chen, Aili
%A Zhang, Haoyu
%A Wu, Libo
%A Yinghui, Xu
%A Xiao, Yanghua
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F zhang-etal-2026-pr
%X Behavioral problems in deployed LLMs are often discovered when only limited supervision is available and repeating full alignment training is impractical. Existing post-hoc methods have complementary limitations: activation steering lacks input adaptivity, while standard LoRA learns without an explicit behavioral direction prior and can be unstable in low-data settings. We introduce PR-LoRA, a direction-regularized LoRA variant that initializes LoRA B matrices with per-layer contrastive behavioral directions and anchors them with an L₂ penalty, while leaving A freely trainable. We evaluate PR-LoRA on seven CAA behaviors across Qwen3 and LLaMA models against activation steering and parameter-efficient adaptation baselines. On Qwen3-8B, PR-LoRA lowers the seven-behavior macro-average undesirable-behavior score from .405 to .374 relative to standard LoRA, a 7.8% relative reduction, and outperforms DPO-LoRA and LoReFT under the tested low-update budget.
%U https://aclanthology.org/2026.emnlp-main.2317/
%P 52640-52654
Markdown (Informal)
[PR-LoRA: Direction-Regularized Low-Rank Adaptation for Behavioral Alignment in Large Language Models](https://aclanthology.org/2026.emnlp-main.2317/) (Zhang et al., EMNLP 2026)
ACL