@inproceedings{yu-etal-2026-stabilizing,
title = "Stabilizing Unsupervised Self-Evolution of {MLLM}s via Continuous Softened Retracing re{S}ampling",
author = "Yu, Yunyao and
Wu, ZhengXian and
Chen, Zhuohong and
Xu, Hangrui and
Liao, Zirui and
Deng, Xiangwen and
Liu, Zhifang and
Shi, Senyuan and
Wang, Haoqian",
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.199/",
pages = "4555--4571",
ISBN = "979-8-89176-459-0",
abstract = "In the unsupervised self-evolution of Multimodal Large Language Models, the quality of feedback signals during post-training is pivotal for stable and effective learning. However, existing self-evolution methods predominantly rely on majority voting to select the most frequent output as the pseudo-golden answer, which may stem from the model{'}s intrinsic biases rather than guaranteeing the objective correctness of the reasoning paths. To counteract the degradation, we propose $\textbf{C}$ontinuous $\textbf{S}$oftened $\textbf{R}$etracing re$\textbf{S}$ampling ($\textbf{CSRS}$) in MLLM self-evolution. Specifically, we introduce a Retracing Re-inference Mechanism ($\textbf{RRM}$) that the model re-inferences from anchor points to expand the exploration of long-tail reasoning paths. Simultaneously, we propose Softened Frequency Reward ($\textbf{SFR}$), which replaces binary rewards with continuous signals, calibrating reward based on the answers' frequency across sampled reasoning sets. Furthermore, incorporated with Visual Semantic Perturbation ($\textbf{VSP}$), CSRS ensures the model prioritizes mathematical logic over visual superficiality. Experimental results demonstrate that CSRS significantly enhances the reasoning performance of Qwen2.5-VL-7B on benchmarks such as MathVision. We achieve state-of-the-art (SOTA) results in unsupervised self-evolution on geometric tasks."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="yu-etal-2026-stabilizing">
<titleInfo>
<title>Stabilizing Unsupervised Self-Evolution of MLLMs via Continuous Softened Retracing reSampling</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yunyao</namePart>
<namePart type="family">Yu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">ZhengXian</namePart>
<namePart type="family">Wu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhuohong</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hangrui</namePart>
<namePart type="family">Xu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zirui</namePart>
<namePart type="family">Liao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xiangwen</namePart>
<namePart type="family">Deng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhifang</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Senyuan</namePart>
<namePart type="family">Shi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Haoqian</namePart>
<namePart type="family">Wang</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>In the unsupervised self-evolution of Multimodal Large Language Models, the quality of feedback signals during post-training is pivotal for stable and effective learning. However, existing self-evolution methods predominantly rely on majority voting to select the most frequent output as the pseudo-golden answer, which may stem from the model’s intrinsic biases rather than guaranteeing the objective correctness of the reasoning paths. To counteract the degradation, we propose Continuous Softened Retracing reSampling (CSRS) in MLLM self-evolution. Specifically, we introduce a Retracing Re-inference Mechanism (RRM) that the model re-inferences from anchor points to expand the exploration of long-tail reasoning paths. Simultaneously, we propose Softened Frequency Reward (SFR), which replaces binary rewards with continuous signals, calibrating reward based on the answers’ frequency across sampled reasoning sets. Furthermore, incorporated with Visual Semantic Perturbation (VSP), CSRS ensures the model prioritizes mathematical logic over visual superficiality. Experimental results demonstrate that CSRS significantly enhances the reasoning performance of Qwen2.5-VL-7B on benchmarks such as MathVision. We achieve state-of-the-art (SOTA) results in unsupervised self-evolution on geometric tasks.</abstract>
<identifier type="citekey">yu-etal-2026-stabilizing</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.199/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>4555</start>
<end>4571</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Stabilizing Unsupervised Self-Evolution of MLLMs via Continuous Softened Retracing reSampling
%A Yu, Yunyao
%A Wu, ZhengXian
%A Chen, Zhuohong
%A Xu, Hangrui
%A Liao, Zirui
%A Deng, Xiangwen
%A Liu, Zhifang
%A Shi, Senyuan
%A Wang, Haoqian
%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 yu-etal-2026-stabilizing
%X In the unsupervised self-evolution of Multimodal Large Language Models, the quality of feedback signals during post-training is pivotal for stable and effective learning. However, existing self-evolution methods predominantly rely on majority voting to select the most frequent output as the pseudo-golden answer, which may stem from the model’s intrinsic biases rather than guaranteeing the objective correctness of the reasoning paths. To counteract the degradation, we propose Continuous Softened Retracing reSampling (CSRS) in MLLM self-evolution. Specifically, we introduce a Retracing Re-inference Mechanism (RRM) that the model re-inferences from anchor points to expand the exploration of long-tail reasoning paths. Simultaneously, we propose Softened Frequency Reward (SFR), which replaces binary rewards with continuous signals, calibrating reward based on the answers’ frequency across sampled reasoning sets. Furthermore, incorporated with Visual Semantic Perturbation (VSP), CSRS ensures the model prioritizes mathematical logic over visual superficiality. Experimental results demonstrate that CSRS significantly enhances the reasoning performance of Qwen2.5-VL-7B on benchmarks such as MathVision. We achieve state-of-the-art (SOTA) results in unsupervised self-evolution on geometric tasks.
%U https://aclanthology.org/2026.emnlp-main.199/
%P 4555-4571
Markdown (Informal)
[Stabilizing Unsupervised Self-Evolution of MLLMs via Continuous Softened Retracing reSampling](https://aclanthology.org/2026.emnlp-main.199/) (Yu et al., EMNLP 2026)
ACL
- Yunyao Yu, ZhengXian Wu, Zhuohong Chen, Hangrui Xu, Zirui Liao, Xiangwen Deng, Zhifang Liu, Senyuan Shi, and Haoqian Wang. 2026. Stabilizing Unsupervised Self-Evolution of MLLMs via Continuous Softened Retracing reSampling. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 4555–4571, Budapest, Hungary. Association for Computational Linguistics.