@inproceedings{hong-etal-2026-roco,
title = "{R}o{C}o-{ACE}: Rollout-Conditioned Online Distillation for Retention-Aware Knowledge Injection",
author = "Hong, Yan and
Li, Wei and
Xiu, Kedong and
Lan, Jun and
Zhou, Shuheng and
Lyu, Zhongcai and
Zhu, Huijia and
Wang, Weiqiang and
Zhang, Jianfu",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.1840/",
pages = "41847--41869",
ISBN = "979-8-89176-459-0",
abstract = "Knowledge injection updates pretrained MLLMs with new factual or domain-specific knowledge, but fitting full authoritative answers can cause drift in non-updated behavior. Online distillation limits this drift by training on model-generated rollouts, yet uniform reference-conditioned distillation gives coarse supervision: it can under-emphasize reference-supported rollout tokens and supervise omitted facts only indirectly. We introduce RoCo-ACE, a rollout-conditioned online distillation objective for knowledge injection: RoCo uses same-rollout reference-free/reference-conditioned likelihood contrast to reallocate additional distillation weight to reference-supported rollout tokens, while ACE adds sparse reference-side anchored correction for authoritative anchors omitted from the rollout, without full-answer imitation. Across three knowledge-injection settings, six retention benchmarks, multiple baselines and base models, RoCo-ACE achieves the best injected-knowledge accuracy among compared methods while keeping evaluated retention close to the base model."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="hong-etal-2026-roco">
<titleInfo>
<title>RoCo-ACE: Rollout-Conditioned Online Distillation for Retention-Aware Knowledge Injection</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yan</namePart>
<namePart type="family">Hong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Wei</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kedong</namePart>
<namePart type="family">Xiu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jun</namePart>
<namePart type="family">Lan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shuheng</namePart>
<namePart type="family">Zhou</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhongcai</namePart>
<namePart type="family">Lyu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Huijia</namePart>
<namePart type="family">Zhu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Weiqiang</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jianfu</namePart>
<namePart type="family">Zhang</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>Knowledge injection updates pretrained MLLMs with new factual or domain-specific knowledge, but fitting full authoritative answers can cause drift in non-updated behavior. Online distillation limits this drift by training on model-generated rollouts, yet uniform reference-conditioned distillation gives coarse supervision: it can under-emphasize reference-supported rollout tokens and supervise omitted facts only indirectly. We introduce RoCo-ACE, a rollout-conditioned online distillation objective for knowledge injection: RoCo uses same-rollout reference-free/reference-conditioned likelihood contrast to reallocate additional distillation weight to reference-supported rollout tokens, while ACE adds sparse reference-side anchored correction for authoritative anchors omitted from the rollout, without full-answer imitation. Across three knowledge-injection settings, six retention benchmarks, multiple baselines and base models, RoCo-ACE achieves the best injected-knowledge accuracy among compared methods while keeping evaluated retention close to the base model.</abstract>
<identifier type="citekey">hong-etal-2026-roco</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.1840/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>41847</start>
<end>41869</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T RoCo-ACE: Rollout-Conditioned Online Distillation for Retention-Aware Knowledge Injection
%A Hong, Yan
%A Li, Wei
%A Xiu, Kedong
%A Lan, Jun
%A Zhou, Shuheng
%A Lyu, Zhongcai
%A Zhu, Huijia
%A Wang, Weiqiang
%A Zhang, Jianfu
%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 hong-etal-2026-roco
%X Knowledge injection updates pretrained MLLMs with new factual or domain-specific knowledge, but fitting full authoritative answers can cause drift in non-updated behavior. Online distillation limits this drift by training on model-generated rollouts, yet uniform reference-conditioned distillation gives coarse supervision: it can under-emphasize reference-supported rollout tokens and supervise omitted facts only indirectly. We introduce RoCo-ACE, a rollout-conditioned online distillation objective for knowledge injection: RoCo uses same-rollout reference-free/reference-conditioned likelihood contrast to reallocate additional distillation weight to reference-supported rollout tokens, while ACE adds sparse reference-side anchored correction for authoritative anchors omitted from the rollout, without full-answer imitation. Across three knowledge-injection settings, six retention benchmarks, multiple baselines and base models, RoCo-ACE achieves the best injected-knowledge accuracy among compared methods while keeping evaluated retention close to the base model.
%U https://aclanthology.org/2026.emnlp-main.1840/
%P 41847-41869
Markdown (Informal)
[RoCo-ACE: Rollout-Conditioned Online Distillation for Retention-Aware Knowledge Injection](https://aclanthology.org/2026.emnlp-main.1840/) (Hong et al., EMNLP 2026)
ACL
- Yan Hong, Wei Li, Kedong Xiu, Jun Lan, Shuheng Zhou, Zhongcai Lyu, Huijia Zhu, Weiqiang Wang, and Jianfu Zhang. 2026. RoCo-ACE: Rollout-Conditioned Online Distillation for Retention-Aware Knowledge Injection. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 41847–41869, Budapest, Hungary. Association for Computational Linguistics.