@inproceedings{wang-etal-2026-charting,
title = "Charting Empirical Patterns for {LLM} Fine-Tuning in Scientific Multi-Discipline Learning",
author = "Wang, Lintao and
Lu, Zhuqiang and
Zhu, Yilin and
Hu, Kun and
Yin, Zhenfei and
Tang, Shixiang and
Wang, Zhiyong and
Ouyang, Wanli and
Ma, Xinzhu",
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.312/",
pages = "7182--7203",
ISBN = "979-8-89176-459-0",
abstract = "While large language models (LLMs) perform strongly after adaptation within individual scientific domains, their learning dynamics in multi-disciplinary contexts remain poorly understood, despite the promise of improved generalization and broader applicability through cross-domain knowledge synergy. We present a systematic study of multi-discipline LLM fine-tuning across five scientific disciplines and compare the learning dynamics of full fine-tuning, LoRA, LoRA-MoE, and LoRA composition. In our naturally imbalanced setting, we show that multi-disciplinary learning is substantially more variable than single-discipline training and distill four empirical patterns: (1) Balance-then-Diversity: increasing scarce-domain exposure helps, but broad source-pool diversity must be preserved; (2) Merge-then-Align: general instruction data restores instruction-following behavior needed for effective joint adaptation; (3) Optimize-then-Scale: increasing trainable capacity offers limited gains without prior design optimization; and (4) Share-then-Specialize: asymmetric LoRA-MoE improves multi-discipline performance through a shared low-rank projection. Together, these patterns form a practical recipe for principled multi-discipline fine-tuning and provide actionable guidance for developing generalizable scientific LLMs."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="wang-etal-2026-charting">
<titleInfo>
<title>Charting Empirical Patterns for LLM Fine-Tuning in Scientific Multi-Discipline Learning</title>
</titleInfo>
<name type="personal">
<namePart type="given">Lintao</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhuqiang</namePart>
<namePart type="family">Lu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yilin</namePart>
<namePart type="family">Zhu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kun</namePart>
<namePart type="family">Hu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhenfei</namePart>
<namePart type="family">Yin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shixiang</namePart>
<namePart type="family">Tang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhiyong</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Wanli</namePart>
<namePart type="family">Ouyang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xinzhu</namePart>
<namePart type="family">Ma</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>While large language models (LLMs) perform strongly after adaptation within individual scientific domains, their learning dynamics in multi-disciplinary contexts remain poorly understood, despite the promise of improved generalization and broader applicability through cross-domain knowledge synergy. We present a systematic study of multi-discipline LLM fine-tuning across five scientific disciplines and compare the learning dynamics of full fine-tuning, LoRA, LoRA-MoE, and LoRA composition. In our naturally imbalanced setting, we show that multi-disciplinary learning is substantially more variable than single-discipline training and distill four empirical patterns: (1) Balance-then-Diversity: increasing scarce-domain exposure helps, but broad source-pool diversity must be preserved; (2) Merge-then-Align: general instruction data restores instruction-following behavior needed for effective joint adaptation; (3) Optimize-then-Scale: increasing trainable capacity offers limited gains without prior design optimization; and (4) Share-then-Specialize: asymmetric LoRA-MoE improves multi-discipline performance through a shared low-rank projection. Together, these patterns form a practical recipe for principled multi-discipline fine-tuning and provide actionable guidance for developing generalizable scientific LLMs.</abstract>
<identifier type="citekey">wang-etal-2026-charting</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.312/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>7182</start>
<end>7203</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Charting Empirical Patterns for LLM Fine-Tuning in Scientific Multi-Discipline Learning
%A Wang, Lintao
%A Lu, Zhuqiang
%A Zhu, Yilin
%A Hu, Kun
%A Yin, Zhenfei
%A Tang, Shixiang
%A Wang, Zhiyong
%A Ouyang, Wanli
%A Ma, Xinzhu
%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 wang-etal-2026-charting
%X While large language models (LLMs) perform strongly after adaptation within individual scientific domains, their learning dynamics in multi-disciplinary contexts remain poorly understood, despite the promise of improved generalization and broader applicability through cross-domain knowledge synergy. We present a systematic study of multi-discipline LLM fine-tuning across five scientific disciplines and compare the learning dynamics of full fine-tuning, LoRA, LoRA-MoE, and LoRA composition. In our naturally imbalanced setting, we show that multi-disciplinary learning is substantially more variable than single-discipline training and distill four empirical patterns: (1) Balance-then-Diversity: increasing scarce-domain exposure helps, but broad source-pool diversity must be preserved; (2) Merge-then-Align: general instruction data restores instruction-following behavior needed for effective joint adaptation; (3) Optimize-then-Scale: increasing trainable capacity offers limited gains without prior design optimization; and (4) Share-then-Specialize: asymmetric LoRA-MoE improves multi-discipline performance through a shared low-rank projection. Together, these patterns form a practical recipe for principled multi-discipline fine-tuning and provide actionable guidance for developing generalizable scientific LLMs.
%U https://aclanthology.org/2026.emnlp-main.312/
%P 7182-7203
Markdown (Informal)
[Charting Empirical Patterns for LLM Fine-Tuning in Scientific Multi-Discipline Learning](https://aclanthology.org/2026.emnlp-main.312/) (Wang et al., EMNLP 2026)
ACL
- Lintao Wang, Zhuqiang Lu, Yilin Zhu, Kun Hu, Zhenfei Yin, Shixiang Tang, Zhiyong Wang, Wanli Ouyang, and Xinzhu Ma. 2026. Charting Empirical Patterns for LLM Fine-Tuning in Scientific Multi-Discipline Learning. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 7182–7203, Budapest, Hungary. Association for Computational Linguistics.