@inproceedings{kappiyath-etal-2026-maple,
title = "{MAPLE}: Masked Adapter Prototype Learning for {OOD} generalization",
author = "Kappiyath, Adarsh and
Chaudhuri, Abhra and
Li, Jiaxi and
Jaiswal, Ajay Kumar and
Wang, Xilu and
Shen, Li and
Kundu, Souvik and
Zhu, Xiatian and
Yin, Lu",
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.148/",
pages = "3419--3432",
ISBN = "979-8-89176-459-0",
abstract = "Parameter-efficient fine-tuning with adapters (e.g., LoRA) equips LLMs with task-specific skills. However, utilizing multiple pretrained adapters for out-of-distribution (OOD) generalization remains challenging. Existing techniques for OOD generalization using multiple pretrained LoRAs, route inputs using LoRA representations (\textit{prototypes}) obtained independently, assuming these representations capture complementary information. However, we observe that for existing methods, in-distribution and OOD routing entropies are often comparable, thus bringing the complementarity assumption into question. We derive the theoretical conditions that could lead to a violation of such assumptions, distilling the cause down to the presence of shared, noisy prototype subspaces. Based on this, we introduce \textbf{MAPLE (Masked-Adapter Prototype LEarning)}, a simple learning framework that refines LoRA prototypes by masking the target task{'}s LoRA during prototype learning. In doing so, it encourages prototypes to discard noisy attributes, improving prototype discriminability, routing, and OOD generalization. Extensive experiments on language models of varying size, such as Phi-2 (2.7B) and LLaMA-3 (8B) equipped with heterogeneous pools of pretrained LoRAs, show that MAPLE improves the LoRA representation and thus achieves state-of-the-art performance across multiple benchmarks."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="kappiyath-etal-2026-maple">
<titleInfo>
<title>MAPLE: Masked Adapter Prototype Learning for OOD generalization</title>
</titleInfo>
<name type="personal">
<namePart type="given">Adarsh</namePart>
<namePart type="family">Kappiyath</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Abhra</namePart>
<namePart type="family">Chaudhuri</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiaxi</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ajay</namePart>
<namePart type="given">Kumar</namePart>
<namePart type="family">Jaiswal</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xilu</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Li</namePart>
<namePart type="family">Shen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Souvik</namePart>
<namePart type="family">Kundu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xiatian</namePart>
<namePart type="family">Zhu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Lu</namePart>
<namePart type="family">Yin</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>Parameter-efficient fine-tuning with adapters (e.g., LoRA) equips LLMs with task-specific skills. However, utilizing multiple pretrained adapters for out-of-distribution (OOD) generalization remains challenging. Existing techniques for OOD generalization using multiple pretrained LoRAs, route inputs using LoRA representations (prototypes) obtained independently, assuming these representations capture complementary information. However, we observe that for existing methods, in-distribution and OOD routing entropies are often comparable, thus bringing the complementarity assumption into question. We derive the theoretical conditions that could lead to a violation of such assumptions, distilling the cause down to the presence of shared, noisy prototype subspaces. Based on this, we introduce MAPLE (Masked-Adapter Prototype LEarning), a simple learning framework that refines LoRA prototypes by masking the target task’s LoRA during prototype learning. In doing so, it encourages prototypes to discard noisy attributes, improving prototype discriminability, routing, and OOD generalization. Extensive experiments on language models of varying size, such as Phi-2 (2.7B) and LLaMA-3 (8B) equipped with heterogeneous pools of pretrained LoRAs, show that MAPLE improves the LoRA representation and thus achieves state-of-the-art performance across multiple benchmarks.</abstract>
<identifier type="citekey">kappiyath-etal-2026-maple</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.148/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>3419</start>
<end>3432</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T MAPLE: Masked Adapter Prototype Learning for OOD generalization
%A Kappiyath, Adarsh
%A Chaudhuri, Abhra
%A Li, Jiaxi
%A Jaiswal, Ajay Kumar
%A Wang, Xilu
%A Shen, Li
%A Kundu, Souvik
%A Zhu, Xiatian
%A Yin, Lu
%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 kappiyath-etal-2026-maple
%X Parameter-efficient fine-tuning with adapters (e.g., LoRA) equips LLMs with task-specific skills. However, utilizing multiple pretrained adapters for out-of-distribution (OOD) generalization remains challenging. Existing techniques for OOD generalization using multiple pretrained LoRAs, route inputs using LoRA representations (prototypes) obtained independently, assuming these representations capture complementary information. However, we observe that for existing methods, in-distribution and OOD routing entropies are often comparable, thus bringing the complementarity assumption into question. We derive the theoretical conditions that could lead to a violation of such assumptions, distilling the cause down to the presence of shared, noisy prototype subspaces. Based on this, we introduce MAPLE (Masked-Adapter Prototype LEarning), a simple learning framework that refines LoRA prototypes by masking the target task’s LoRA during prototype learning. In doing so, it encourages prototypes to discard noisy attributes, improving prototype discriminability, routing, and OOD generalization. Extensive experiments on language models of varying size, such as Phi-2 (2.7B) and LLaMA-3 (8B) equipped with heterogeneous pools of pretrained LoRAs, show that MAPLE improves the LoRA representation and thus achieves state-of-the-art performance across multiple benchmarks.
%U https://aclanthology.org/2026.emnlp-main.148/
%P 3419-3432
Markdown (Informal)
[MAPLE: Masked Adapter Prototype Learning for OOD generalization](https://aclanthology.org/2026.emnlp-main.148/) (Kappiyath et al., EMNLP 2026)
ACL
- Adarsh Kappiyath, Abhra Chaudhuri, Jiaxi Li, Ajay Kumar Jaiswal, Xilu Wang, Li Shen, Souvik Kundu, Xiatian Zhu, and Lu Yin. 2026. MAPLE: Masked Adapter Prototype Learning for OOD generalization. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 3419–3432, Budapest, Hungary. Association for Computational Linguistics.