<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="kim-etal-2026-pivot">
    <titleInfo>
        <title>Pivot-SD: Efficient Self-Distillation for Masked Diffusion Language Models</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Seo</namePart>
        <namePart type="given">Hyun</namePart>
        <namePart type="family">Kim</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Sunwoo</namePart>
        <namePart type="family">Hong</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Younwoo</namePart>
        <namePart type="family">Choi</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Chen-Hao</namePart>
        <namePart type="family">Chao</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Se-Young</namePart>
        <namePart type="family">Yun</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Rahul</namePart>
        <namePart type="given">G</namePart>
        <namePart type="family">Krishnan</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>Masked diffusion language models (dLMs) offer a promising parallel alternative to autoregressive models for complex reasoning. However, they face a unique credit-assignment challenge: reasoning failures typically stem from a few critical, early commitments that cascade downstream. Existing post-training overlooks this dynamic: supervised fine-tuning (SFT) weights all tokens uniformly, while online reinforcement learning (RL) attributes a single trajectory-level reward to every position. We introduce Pivot-SD, an efficient offline framework supervising only these high-impact bottlenecks (pivots). Pivot-SD isolates pivots using an information-gain metric measuring uncertainty reduction over the remaining masked block. It applies bidirectional supervision: successful pivots are reinforced with cross-entropy, while critical errors in failed trajectories are suppressed via targeted unlikelihood, preserving valid surrounding sub-steps. Using only 200 questions and four rollouts each, Pivot-SD improves LLaDA-8B-Instruct over full-sequence SFT and budget-matched diffusion RL baselines across math and code benchmarks.</abstract>
    <identifier type="citekey">kim-etal-2026-pivot</identifier>
    <location>
        <url>https://aclanthology.org/2026.emnlp-main.2146/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>48790</start>
            <end>48806</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Pivot-SD: Efficient Self-Distillation for Masked Diffusion Language Models
%A Kim, Seo Hyun
%A Hong, Sunwoo
%A Choi, Younwoo
%A Chao, Chen-Hao
%A Yun, Se-Young
%A Krishnan, Rahul G.
%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 kim-etal-2026-pivot
%X Masked diffusion language models (dLMs) offer a promising parallel alternative to autoregressive models for complex reasoning. However, they face a unique credit-assignment challenge: reasoning failures typically stem from a few critical, early commitments that cascade downstream. Existing post-training overlooks this dynamic: supervised fine-tuning (SFT) weights all tokens uniformly, while online reinforcement learning (RL) attributes a single trajectory-level reward to every position. We introduce Pivot-SD, an efficient offline framework supervising only these high-impact bottlenecks (pivots). Pivot-SD isolates pivots using an information-gain metric measuring uncertainty reduction over the remaining masked block. It applies bidirectional supervision: successful pivots are reinforced with cross-entropy, while critical errors in failed trajectories are suppressed via targeted unlikelihood, preserving valid surrounding sub-steps. Using only 200 questions and four rollouts each, Pivot-SD improves LLaDA-8B-Instruct over full-sequence SFT and budget-matched diffusion RL baselines across math and code benchmarks.
%U https://aclanthology.org/2026.emnlp-main.2146/
%P 48790-48806

Markdown (Informal)

[Pivot-SD: Efficient Self-Distillation for Masked Diffusion Language Models](https://aclanthology.org/2026.emnlp-main.2146/) (Kim et al., EMNLP 2026)

• Pivot-SD: Efficient Self-Distillation for Masked Diffusion Language Models (Kim et al., EMNLP 2026)

ACL

• Seo Hyun Kim, Sunwoo Hong, Younwoo Choi, Chen-Hao Chao, Se-Young Yun, and Rahul G Krishnan. 2026. Pivot-SD: Efficient Self-Distillation for Masked Diffusion Language Models. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 48790–48806, Budapest, Hungary. Association for Computational Linguistics.