<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="han-etal-2026-s4r">
    <titleInfo>
        <title>S⁴R: Selective Sampling, Subspaces, and Sparse Reconstruction for Compressed Long-Context KV Caching</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Jialong</namePart>
        <namePart type="family">Han</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">You</namePart>
        <namePart type="family">Wu</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Kewei</namePart>
        <namePart type="family">Tu</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <originInfo>
        <dateIssued>2026-11</dateIssued>
    </originInfo>
    <typeOfResource>text</typeOfResource>
    <relatedItem type="host">
        <titleInfo>
            <title>Proceedings of the 15th International Joint Conference on Natural Language Processing and the 5th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (Volume 1: Long Papers)</title>
        </titleInfo>
        <name type="personal">
            <namePart type="given">Yue</namePart>
            <namePart type="family">Zhang</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Ruifeng</namePart>
            <namePart type="family">Xu</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Ayu</namePart>
            <namePart type="family">Purwarianti</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Tong</namePart>
            <namePart type="family">Rong</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Jun</namePart>
            <namePart type="family">Suzuki</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <originInfo>
            <publisher>Association for Computational Linguistics</publisher>
            <place>
                <placeTerm type="text">Hengqin, China</placeTerm>
            </place>
        </originInfo>
        <genre authority="marcgt">conference publication</genre>
        <identifier type="isbn">979-8-89176-439-2</identifier>
    </relatedItem>
    <abstract>The growth of context window lengths in Large Language Models (LLMs) significantly enhances their long-context capabilities but incurs prohibitive memory costs due to the Key-Value (KV) cache. Although low-rank compression of KV cache is a promising remedy, existing methods face a dilemma: offline approaches depend on external calibration data, whereas online approaches incur substantial compute for full-prompt decomposition and reconstruction. In this paper, we propose **S⁴R**, which builds low-rank subspaces from selectively sampled tokens and computes attention over a sparsely reconstructed KV representation. S⁴R uses prompt-aware initialization to build initial key/value bases from a representative prompt subset, trading off calibration-data dependence against prefilling cost. Because fully reconstructing the cache at every decoding step is prohibitively expensive and hurts throughput, we further adopt sparse reconstruction to retain only informative positions during decoding. Extensive experiments on LongBench and RULER with Llama and Qwen model families show that S⁴R achieves up to 5x KV compression with near full-cache accuracy, combining the efficiency of fixed compression with the adaptability of prompt-dependent methods. Our code is publicly available at https://github.com/Draconis98/S4R</abstract>
    <identifier type="citekey">han-etal-2026-s4r</identifier>
    <location>
        <url>https://aclanthology.org/2026.aacl-long.137/</url>
    </location>
    <part>
        <date>2026-11</date>
        <extent unit="page">
            <start>2997</start>
            <end>3017</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T S⁴R: Selective Sampling, Subspaces, and Sparse Reconstruction for Compressed Long-Context KV Caching
%A Han, Jialong
%A Wu, You
%A Tu, Kewei
%Y Zhang, Yue
%Y Xu, Ruifeng
%Y Purwarianti, Ayu
%Y Rong, Tong
%Y Suzuki, Jun
%S Proceedings of the 15th International Joint Conference on Natural Language Processing and the 5th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (Volume 1: Long Papers)
%D 2026
%8 November
%I Association for Computational Linguistics
%C Hengqin, China
%@ 979-8-89176-439-2
%F han-etal-2026-s4r
%X The growth of context window lengths in Large Language Models (LLMs) significantly enhances their long-context capabilities but incurs prohibitive memory costs due to the Key-Value (KV) cache. Although low-rank compression of KV cache is a promising remedy, existing methods face a dilemma: offline approaches depend on external calibration data, whereas online approaches incur substantial compute for full-prompt decomposition and reconstruction. In this paper, we propose **S⁴R**, which builds low-rank subspaces from selectively sampled tokens and computes attention over a sparsely reconstructed KV representation. S⁴R uses prompt-aware initialization to build initial key/value bases from a representative prompt subset, trading off calibration-data dependence against prefilling cost. Because fully reconstructing the cache at every decoding step is prohibitively expensive and hurts throughput, we further adopt sparse reconstruction to retain only informative positions during decoding. Extensive experiments on LongBench and RULER with Llama and Qwen model families show that S⁴R achieves up to 5x KV compression with near full-cache accuracy, combining the efficiency of fixed compression with the adaptability of prompt-dependent methods. Our code is publicly available at https://github.com/Draconis98/S4R
%U https://aclanthology.org/2026.aacl-long.137/
%P 2997-3017

Markdown (Informal)

[S4R: Selective Sampling, Subspaces, and Sparse Reconstruction for Compressed Long-Context KV Caching](https://aclanthology.org/2026.aacl-long.137/) (Han et al., AACL-IJCNLP 2026)

• S4R: Selective Sampling, Subspaces, and Sparse Reconstruction for Compressed Long-Context KV Caching (Han et al., AACL-IJCNLP 2026)

ACL

• Jialong Han, You Wu, and Kewei Tu. 2026. S4R: Selective Sampling, Subspaces, and Sparse Reconstruction for Compressed Long-Context KV Caching. In Proceedings of the 15th International Joint Conference on Natural Language Processing and the 5th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (Volume 1: Long Papers), pages 2997–3017, Hengqin, China. Association for Computational Linguistics.