@inproceedings{kawarada-etal-2026-efficient,
title = "Efficient Many-Shot In-Context Learning with Task-Level Compression and Instance-Level Retrieval",
author = "Kawarada, Masayuki and
Honda, Ukyo and
Takase, Sho",
editor = "Zhang, Yue and
Xu, Ruifeng and
Purwarianti, Ayu and
Rong, Tong and
Suzuki, Jun",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {IJCNLP}-{AACL} 2026",
month = nov,
year = "2026",
address = "Hengqin, China",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-aacl.179/",
pages = "3547--3555",
ISBN = "979-8-89176-444-6",
abstract = "Many-shot in-context learning (ICL) lets large language models exploit hundreds or thousands of demonstrations, but its cost grows with the prompt. We argue that many-shot demonstrations carry two kinds of information: task-level information, such as task structure, which emerges only across many examples, and instance-level information, such as exact input-output associations and rare exceptions, which resides in individual demonstrations. Existing efficiency methods commit all information to a single placement: demonstration retrieval keeps a few demonstrations as raw text, while context compression places the full pool into a compact memory. We study where each kind of information should be placed. On ManyICLBench, each one-sided method falls short on tasks that depend on the information it discards, while combining the two retains much of the full many-shot performance at a much smaller inference budget. Synthetic tasks further show that compressed memory and retrieved examples can complement each other but sometimes interfere."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="kawarada-etal-2026-efficient">
<titleInfo>
<title>Efficient Many-Shot In-Context Learning with Task-Level Compression and Instance-Level Retrieval</title>
</titleInfo>
<name type="personal">
<namePart type="given">Masayuki</namePart>
<namePart type="family">Kawarada</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ukyo</namePart>
<namePart type="family">Honda</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sho</namePart>
<namePart type="family">Takase</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>Findings of the Association for Computational Linguistics: IJCNLP-AACL 2026</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">979-8-89176-444-6</identifier>
</relatedItem>
<abstract>Many-shot in-context learning (ICL) lets large language models exploit hundreds or thousands of demonstrations, but its cost grows with the prompt. We argue that many-shot demonstrations carry two kinds of information: task-level information, such as task structure, which emerges only across many examples, and instance-level information, such as exact input-output associations and rare exceptions, which resides in individual demonstrations. Existing efficiency methods commit all information to a single placement: demonstration retrieval keeps a few demonstrations as raw text, while context compression places the full pool into a compact memory. We study where each kind of information should be placed. On ManyICLBench, each one-sided method falls short on tasks that depend on the information it discards, while combining the two retains much of the full many-shot performance at a much smaller inference budget. Synthetic tasks further show that compressed memory and retrieved examples can complement each other but sometimes interfere.</abstract>
<identifier type="citekey">kawarada-etal-2026-efficient</identifier>
<location>
<url>https://aclanthology.org/2026.findings-aacl.179/</url>
</location>
<part>
<date>2026-11</date>
<extent unit="page">
<start>3547</start>
<end>3555</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Efficient Many-Shot In-Context Learning with Task-Level Compression and Instance-Level Retrieval
%A Kawarada, Masayuki
%A Honda, Ukyo
%A Takase, Sho
%Y Zhang, Yue
%Y Xu, Ruifeng
%Y Purwarianti, Ayu
%Y Rong, Tong
%Y Suzuki, Jun
%S Findings of the Association for Computational Linguistics: IJCNLP-AACL 2026
%D 2026
%8 November
%I Association for Computational Linguistics
%C Hengqin, China
%@ 979-8-89176-444-6
%F kawarada-etal-2026-efficient
%X Many-shot in-context learning (ICL) lets large language models exploit hundreds or thousands of demonstrations, but its cost grows with the prompt. We argue that many-shot demonstrations carry two kinds of information: task-level information, such as task structure, which emerges only across many examples, and instance-level information, such as exact input-output associations and rare exceptions, which resides in individual demonstrations. Existing efficiency methods commit all information to a single placement: demonstration retrieval keeps a few demonstrations as raw text, while context compression places the full pool into a compact memory. We study where each kind of information should be placed. On ManyICLBench, each one-sided method falls short on tasks that depend on the information it discards, while combining the two retains much of the full many-shot performance at a much smaller inference budget. Synthetic tasks further show that compressed memory and retrieved examples can complement each other but sometimes interfere.
%U https://aclanthology.org/2026.findings-aacl.179/
%P 3547-3555
Markdown (Informal)
[Efficient Many-Shot In-Context Learning with Task-Level Compression and Instance-Level Retrieval](https://aclanthology.org/2026.findings-aacl.179/) (Kawarada et al., Findings 2026)
ACL