@inproceedings{he-etal-2026-fepo,
title = "{FEPO}: Future-Entropy Policy Optimization with Selective Update Modulation",
author = "He, Cunjie and
Cheng, Tianhao and
Li, Zhiqi and
Gu, Jiaxin and
Huang, Xiaoshuang and
Cai, Jiayin and
Jiang, Xiaolong and
Hu, Yao and
Xu, Jing and
Qi, Yuan and
Xu, Zenglin",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-emnlp.1792/",
pages = "38326--38343",
ISBN = "979-8-89176-462-0",
abstract = "Existing RL methods for language models typically use token-level entropy, which captures uncertainty at the current position rather than over future generation. We study future entropy, i.e., uncertainty over the remaining suffix conditioned on the current prefix. By the entropy chain rule, local entropy is the expected decrease in future entropy, while the realized decrease depends on the chosen token, yielding an action-dependent future-entropy gap. We evaluate this signal at inference time with MIN-F-MC, an oracle-style probe that selects the candidate token with the lowest estimated future-entropy rate at high-entropy positions. On mathematical reasoning tasks, MIN-F-MC yields higher mean accuracy than greedy and Random TopK across the tested checkpoints and datasets. We then propose FEPO (Future Entropy Policy Optimization), which uses realized future-entropy rates at a sparse set of high-entropy positions to scale update magnitudes. Across the tested model{--}benchmark pairs, FEPO improves Mean@32 over GRPO."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="he-etal-2026-fepo">
<titleInfo>
<title>FEPO: Future-Entropy Policy Optimization with Selective Update Modulation</title>
</titleInfo>
<name type="personal">
<namePart type="given">Cunjie</namePart>
<namePart type="family">He</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tianhao</namePart>
<namePart type="family">Cheng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhiqi</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiaxin</namePart>
<namePart type="family">Gu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xiaoshuang</namePart>
<namePart type="family">Huang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiayin</namePart>
<namePart type="family">Cai</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xiaolong</namePart>
<namePart type="family">Jiang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yao</namePart>
<namePart type="family">Hu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jing</namePart>
<namePart type="family">Xu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yuan</namePart>
<namePart type="family">Qi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zenglin</namePart>
<namePart type="family">Xu</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>Existing RL methods for language models typically use token-level entropy, which captures uncertainty at the current position rather than over future generation. We study future entropy, i.e., uncertainty over the remaining suffix conditioned on the current prefix. By the entropy chain rule, local entropy is the expected decrease in future entropy, while the realized decrease depends on the chosen token, yielding an action-dependent future-entropy gap. We evaluate this signal at inference time with MIN-F-MC, an oracle-style probe that selects the candidate token with the lowest estimated future-entropy rate at high-entropy positions. On mathematical reasoning tasks, MIN-F-MC yields higher mean accuracy than greedy and Random TopK across the tested checkpoints and datasets. We then propose FEPO (Future Entropy Policy Optimization), which uses realized future-entropy rates at a sparse set of high-entropy positions to scale update magnitudes. Across the tested model–benchmark pairs, FEPO improves Mean@32 over GRPO.</abstract>
<identifier type="citekey">he-etal-2026-fepo</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.1792/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>38326</start>
<end>38343</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T FEPO: Future-Entropy Policy Optimization with Selective Update Modulation
%A He, Cunjie
%A Cheng, Tianhao
%A Li, Zhiqi
%A Gu, Jiaxin
%A Huang, Xiaoshuang
%A Cai, Jiayin
%A Jiang, Xiaolong
%A Hu, Yao
%A Xu, Jing
%A Qi, Yuan
%A Xu, Zenglin
%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 he-etal-2026-fepo
%X Existing RL methods for language models typically use token-level entropy, which captures uncertainty at the current position rather than over future generation. We study future entropy, i.e., uncertainty over the remaining suffix conditioned on the current prefix. By the entropy chain rule, local entropy is the expected decrease in future entropy, while the realized decrease depends on the chosen token, yielding an action-dependent future-entropy gap. We evaluate this signal at inference time with MIN-F-MC, an oracle-style probe that selects the candidate token with the lowest estimated future-entropy rate at high-entropy positions. On mathematical reasoning tasks, MIN-F-MC yields higher mean accuracy than greedy and Random TopK across the tested checkpoints and datasets. We then propose FEPO (Future Entropy Policy Optimization), which uses realized future-entropy rates at a sparse set of high-entropy positions to scale update magnitudes. Across the tested model–benchmark pairs, FEPO improves Mean@32 over GRPO.
%U https://aclanthology.org/2026.findings-emnlp.1792/
%P 38326-38343
Markdown (Informal)
[FEPO: Future-Entropy Policy Optimization with Selective Update Modulation](https://aclanthology.org/2026.findings-emnlp.1792/) (He et al., Findings 2026)
ACL
- Cunjie He, Tianhao Cheng, Zhiqi Li, Jiaxin Gu, Xiaoshuang Huang, Jiayin Cai, Xiaolong Jiang, Yao Hu, Jing Xu, Yuan Qi, and Zenglin Xu. 2026. FEPO: Future-Entropy Policy Optimization with Selective Update Modulation. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 38326–38343, Budapest, Hungary. Association for Computational Linguistics.