@inproceedings{song-etal-2026-hlora,
title = "{HL}o{RA}: Hierarchical Regularization on Combined Parameter Space for {L}o{RA} Fine-Tuning",
author = "Song, Shezheng and
Xu, Hao and
Peng, Long and
Li, Xiaopeng and
Li, Shasha and
Yu, Jie",
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.1104/",
pages = "25301--25316",
ISBN = "979-8-89176-459-0",
abstract = "Low-Rank Adaptation (LoRA) is widely used for fine-tuning large language models, but suffers from catastrophic forgetting. Existing continual learning methods mitigate this by estimating per-parameter importance independently. However, this assumption breaks down under LoRA{'}s factorization $\Delta W = BA$: infinitely many pairs $(B, A)$ yield the same weight update, making importance scores on $A$ and $B$ separately unreliable. We propose HLoRA, which computes path-integral importance in the combined weight update space, bypassing the unreliable separate estimation on $A$ and $B$. HLoRA applies hierarchical regularization at two levels: element-wise constraints preserve parameters critical for general knowledge, and layer-wise coefficients adapt regularization strength based on each layer{'}s aggregate importance. Experiments on scientific, medical, and physical tasks using GPT-J and LLaMA-3 show that HLoRA mitigates catastrophic forgetting while enhancing adaptability, running approximately 20x faster and requiring only 10{\%}{--}15{\%} of the storage compared to prior methods. Code is released on https://anonymous.4open.science/r/HLoRA-915F."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="song-etal-2026-hlora">
<titleInfo>
<title>HLoRA: Hierarchical Regularization on Combined Parameter Space for LoRA Fine-Tuning</title>
</titleInfo>
<name type="personal">
<namePart type="given">Shezheng</namePart>
<namePart type="family">Song</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hao</namePart>
<namePart type="family">Xu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Long</namePart>
<namePart type="family">Peng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xiaopeng</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shasha</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jie</namePart>
<namePart type="family">Yu</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>Low-Rank Adaptation (LoRA) is widely used for fine-tuning large language models, but suffers from catastrophic forgetting. Existing continual learning methods mitigate this by estimating per-parameter importance independently. However, this assumption breaks down under LoRA’s factorization Δ W = BA: infinitely many pairs (B, A) yield the same weight update, making importance scores on A and B separately unreliable. We propose HLoRA, which computes path-integral importance in the combined weight update space, bypassing the unreliable separate estimation on A and B. HLoRA applies hierarchical regularization at two levels: element-wise constraints preserve parameters critical for general knowledge, and layer-wise coefficients adapt regularization strength based on each layer’s aggregate importance. Experiments on scientific, medical, and physical tasks using GPT-J and LLaMA-3 show that HLoRA mitigates catastrophic forgetting while enhancing adaptability, running approximately 20x faster and requiring only 10%–15% of the storage compared to prior methods. Code is released on https://anonymous.4open.science/r/HLoRA-915F.</abstract>
<identifier type="citekey">song-etal-2026-hlora</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.1104/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>25301</start>
<end>25316</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T HLoRA: Hierarchical Regularization on Combined Parameter Space for LoRA Fine-Tuning
%A Song, Shezheng
%A Xu, Hao
%A Peng, Long
%A Li, Xiaopeng
%A Li, Shasha
%A Yu, Jie
%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 song-etal-2026-hlora
%X Low-Rank Adaptation (LoRA) is widely used for fine-tuning large language models, but suffers from catastrophic forgetting. Existing continual learning methods mitigate this by estimating per-parameter importance independently. However, this assumption breaks down under LoRA’s factorization Δ W = BA: infinitely many pairs (B, A) yield the same weight update, making importance scores on A and B separately unreliable. We propose HLoRA, which computes path-integral importance in the combined weight update space, bypassing the unreliable separate estimation on A and B. HLoRA applies hierarchical regularization at two levels: element-wise constraints preserve parameters critical for general knowledge, and layer-wise coefficients adapt regularization strength based on each layer’s aggregate importance. Experiments on scientific, medical, and physical tasks using GPT-J and LLaMA-3 show that HLoRA mitigates catastrophic forgetting while enhancing adaptability, running approximately 20x faster and requiring only 10%–15% of the storage compared to prior methods. Code is released on https://anonymous.4open.science/r/HLoRA-915F.
%U https://aclanthology.org/2026.emnlp-main.1104/
%P 25301-25316
Markdown (Informal)
[HLoRA: Hierarchical Regularization on Combined Parameter Space for LoRA Fine-Tuning](https://aclanthology.org/2026.emnlp-main.1104/) (Song et al., EMNLP 2026)
ACL