@inproceedings{jang-etal-2026-subt,
title = "{S}ub{T}: Subspace Tuning for Few-shot Generalization of Audio-Language Models",
author = "Jang, Jaehyuk and
Ko, Kangwook and
Lee, Wonjun and
Kim, Changick",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-emnlp.1675/",
pages = "35754--35770",
ISBN = "979-8-89176-462-0",
abstract = "Few-shot parameter-efficient adaptation of pretrained Audio{--}Language Models (ALMs) often improves seen-class performance at the cost of unseen-class generalization, leading to the base-to-new trade-off. We study this failure through zero-shot drift in the text embedding space: few-shot tuning can distort inter-class structure and move adapted embeddings away from their pretrained anchors. We therefore propose Subspace Tuning (SubT), an embedding-space adaptation method that combines a geometry-aware shared transformation with anchoring to the zero-shot prototypes. The learned transformation is transferred to unseen classes, with subspace-aware gating to mitigate negative transfer. Across 11 audio benchmarks, SubT achieves strong few-shot generalization while operating directly on precomputed text embeddings without text-encoder backpropagation. Code is available at https://github.com/jhyukjang/SubT."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="jang-etal-2026-subt">
<titleInfo>
<title>SubT: Subspace Tuning for Few-shot Generalization of Audio-Language Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">Jaehyuk</namePart>
<namePart type="family">Jang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kangwook</namePart>
<namePart type="family">Ko</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Wonjun</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Changick</namePart>
<namePart type="family">Kim</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>Findings of the Association for Computational Linguistics: EMNLP 2026</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">979-8-89176-462-0</identifier>
</relatedItem>
<abstract>Few-shot parameter-efficient adaptation of pretrained Audio–Language Models (ALMs) often improves seen-class performance at the cost of unseen-class generalization, leading to the base-to-new trade-off. We study this failure through zero-shot drift in the text embedding space: few-shot tuning can distort inter-class structure and move adapted embeddings away from their pretrained anchors. We therefore propose Subspace Tuning (SubT), an embedding-space adaptation method that combines a geometry-aware shared transformation with anchoring to the zero-shot prototypes. The learned transformation is transferred to unseen classes, with subspace-aware gating to mitigate negative transfer. Across 11 audio benchmarks, SubT achieves strong few-shot generalization while operating directly on precomputed text embeddings without text-encoder backpropagation. Code is available at https://github.com/jhyukjang/SubT.</abstract>
<identifier type="citekey">jang-etal-2026-subt</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.1675/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>35754</start>
<end>35770</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T SubT: Subspace Tuning for Few-shot Generalization of Audio-Language Models
%A Jang, Jaehyuk
%A Ko, Kangwook
%A Lee, Wonjun
%A Kim, Changick
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F jang-etal-2026-subt
%X Few-shot parameter-efficient adaptation of pretrained Audio–Language Models (ALMs) often improves seen-class performance at the cost of unseen-class generalization, leading to the base-to-new trade-off. We study this failure through zero-shot drift in the text embedding space: few-shot tuning can distort inter-class structure and move adapted embeddings away from their pretrained anchors. We therefore propose Subspace Tuning (SubT), an embedding-space adaptation method that combines a geometry-aware shared transformation with anchoring to the zero-shot prototypes. The learned transformation is transferred to unseen classes, with subspace-aware gating to mitigate negative transfer. Across 11 audio benchmarks, SubT achieves strong few-shot generalization while operating directly on precomputed text embeddings without text-encoder backpropagation. Code is available at https://github.com/jhyukjang/SubT.
%U https://aclanthology.org/2026.findings-emnlp.1675/
%P 35754-35770
Markdown (Informal)
[SubT: Subspace Tuning for Few-shot Generalization of Audio-Language Models](https://aclanthology.org/2026.findings-emnlp.1675/) (Jang et al., Findings 2026)
ACL