<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="varangot-reille-etal-2026-targeting">
    <titleInfo>
        <title>Targeting the Exceptions: Online Active Learning for Retrieval-Based LLM Routers</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Clovis</namePart>
        <namePart type="family">Varangot-Reille</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Antoine</namePart>
        <namePart type="family">Gourru</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Christophe</namePart>
        <namePart type="family">Bouvard</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Baptiste</namePart>
        <namePart type="family">Jeudy</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>Large language Models (LLM) routing reduces inference cost by selecting the most cost-effective model per query, but production deployment requires a router that adapts to an evolving model pool while adding negligible per-query overhead. Unsupervised retrieval-based routers satisfy both, but their support corpus must be built online from streaming traffic, and labeling each query requires running every candidate plus an LLM judge: prohibitively expensive at scale. We frame online corpus construction as stream-based active learning, exploiting a structural property of routing benchmarks: one or two models are globally near-optimal on the majority of queries, so the labeling budget should concentrate on the rare exception regions where selecting a specialist matters. We formalize this with a performance sparsity score, defined as the Kullback-Leibler divergence of a query’s per-candidate performance vector from uniform: high when performance concentrates on a few models and low when all candidates are equivalent. Since this score depends on performances unavailable at inference time, we predict it online from query embeddings with a lightweight regressor. Across four benchmarks, our method SparseNN approaches full-annotation routing performance with up to 44% fewer annotations in the stationary setting and 73% under topic shift. The results transfer to a fully held-out benchmark. A partial model annotation variant achieves higher cost savings, especially on large pools. Finally, both the corpus and the online estimator prove robust to model-pool changes.</abstract>
    <identifier type="citekey">varangot-reille-etal-2026-targeting</identifier>
    <identifier type="doi">10.18653/2026.findings-emnlp.2428</identifier>
    <location>
        <url>https://aclanthology.org/2026.findings-emnlp.2428/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>52117</start>
            <end>52135</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Targeting the Exceptions: Online Active Learning for Retrieval-Based LLM Routers
%A Varangot-Reille, Clovis
%A Gourru, Antoine
%A Bouvard, Christophe
%A Jeudy, Baptiste
%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 varangot-reille-etal-2026-targeting
%X Large language Models (LLM) routing reduces inference cost by selecting the most cost-effective model per query, but production deployment requires a router that adapts to an evolving model pool while adding negligible per-query overhead. Unsupervised retrieval-based routers satisfy both, but their support corpus must be built online from streaming traffic, and labeling each query requires running every candidate plus an LLM judge: prohibitively expensive at scale. We frame online corpus construction as stream-based active learning, exploiting a structural property of routing benchmarks: one or two models are globally near-optimal on the majority of queries, so the labeling budget should concentrate on the rare exception regions where selecting a specialist matters. We formalize this with a performance sparsity score, defined as the Kullback-Leibler divergence of a query’s per-candidate performance vector from uniform: high when performance concentrates on a few models and low when all candidates are equivalent. Since this score depends on performances unavailable at inference time, we predict it online from query embeddings with a lightweight regressor. Across four benchmarks, our method SparseNN approaches full-annotation routing performance with up to 44% fewer annotations in the stationary setting and 73% under topic shift. The results transfer to a fully held-out benchmark. A partial model annotation variant achieves higher cost savings, especially on large pools. Finally, both the corpus and the online estimator prove robust to model-pool changes.
%R 10.18653/2026.findings-emnlp.2428
%U https://aclanthology.org/2026.findings-emnlp.2428/
%U https://doi.org/10.18653/2026.findings-emnlp.2428
%P 52117-52135

Markdown (Informal)

[Targeting the Exceptions: Online Active Learning for Retrieval-Based LLM Routers](https://aclanthology.org/2026.findings-emnlp.2428/) (Varangot-Reille et al., Findings 2026)

• Targeting the Exceptions: Online Active Learning for Retrieval-Based LLM Routers (Varangot-Reille et al., Findings 2026)

ACL

• Clovis Varangot-Reille, Antoine Gourru, Christophe Bouvard, and Baptiste Jeudy. 2026. Targeting the Exceptions: Online Active Learning for Retrieval-Based LLM Routers. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 52117–52135, Budapest, Hungary. Association for Computational Linguistics.