@inproceedings{lin-etal-2026-full-distribution,
title = "Full-Distribution Self-Distillation for Optimizing Soft Latent Reasoning",
author = "Lin, Ziyong and
Wang, Yanting and
Jia, Zixia and
Zheng, Zilong",
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.1972/",
pages = "42061--42073",
ISBN = "979-8-89176-462-0",
abstract = "Soft latent reasoning extends chain-of-thought reasoning beyond discrete token sequences by using continuous concept tokens. However, effectively training models in this space remains challenging: conventional reinforcement learning methods mainly optimize sampled rollout-action likelihoods and therefore underuse the full distributions produced during soft rollouts. We propose **Full-Distribution Self-Distillation (FDSD)**, an off-policy training framework that treats stochastic soft rollout distributions as dense pseudo-labels. FDSD aligns the current model distribution with advantage-weighted rollout targets via KL minimization and importance correction, enabling the model to learn from the full rollout distribution rather than only from sampled-action likelihoods. Experiments on GSM8K, MATH-500, OlympiadBench, and out-of-domain benchmarks show that FDSD achieves consistent average improvements in mathematical reasoning across model scales while preserving OOD generalization, with only modest computational overhead. These results suggest that full-distribution alignment is a promising training direction for soft latent reasoning."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="lin-etal-2026-full-distribution">
<titleInfo>
<title>Full-Distribution Self-Distillation for Optimizing Soft Latent Reasoning</title>
</titleInfo>
<name type="personal">
<namePart type="given">Ziyong</namePart>
<namePart type="family">Lin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yanting</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zixia</namePart>
<namePart type="family">Jia</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zilong</namePart>
<namePart type="family">Zheng</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>Soft latent reasoning extends chain-of-thought reasoning beyond discrete token sequences by using continuous concept tokens. However, effectively training models in this space remains challenging: conventional reinforcement learning methods mainly optimize sampled rollout-action likelihoods and therefore underuse the full distributions produced during soft rollouts. We propose **Full-Distribution Self-Distillation (FDSD)**, an off-policy training framework that treats stochastic soft rollout distributions as dense pseudo-labels. FDSD aligns the current model distribution with advantage-weighted rollout targets via KL minimization and importance correction, enabling the model to learn from the full rollout distribution rather than only from sampled-action likelihoods. Experiments on GSM8K, MATH-500, OlympiadBench, and out-of-domain benchmarks show that FDSD achieves consistent average improvements in mathematical reasoning across model scales while preserving OOD generalization, with only modest computational overhead. These results suggest that full-distribution alignment is a promising training direction for soft latent reasoning.</abstract>
<identifier type="citekey">lin-etal-2026-full-distribution</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.1972/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>42061</start>
<end>42073</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Full-Distribution Self-Distillation for Optimizing Soft Latent Reasoning
%A Lin, Ziyong
%A Wang, Yanting
%A Jia, Zixia
%A Zheng, Zilong
%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 lin-etal-2026-full-distribution
%X Soft latent reasoning extends chain-of-thought reasoning beyond discrete token sequences by using continuous concept tokens. However, effectively training models in this space remains challenging: conventional reinforcement learning methods mainly optimize sampled rollout-action likelihoods and therefore underuse the full distributions produced during soft rollouts. We propose **Full-Distribution Self-Distillation (FDSD)**, an off-policy training framework that treats stochastic soft rollout distributions as dense pseudo-labels. FDSD aligns the current model distribution with advantage-weighted rollout targets via KL minimization and importance correction, enabling the model to learn from the full rollout distribution rather than only from sampled-action likelihoods. Experiments on GSM8K, MATH-500, OlympiadBench, and out-of-domain benchmarks show that FDSD achieves consistent average improvements in mathematical reasoning across model scales while preserving OOD generalization, with only modest computational overhead. These results suggest that full-distribution alignment is a promising training direction for soft latent reasoning.
%U https://aclanthology.org/2026.findings-emnlp.1972/
%P 42061-42073
Markdown (Informal)
[Full-Distribution Self-Distillation for Optimizing Soft Latent Reasoning](https://aclanthology.org/2026.findings-emnlp.1972/) (Lin et al., Findings 2026)
ACL