@inproceedings{jung-etal-2026-breaking,
title = "Breaking Failure Cascades: Step-Aware Reinforcement Learning for Medical Multimodal Reasoning",
author = "Jung, Junha and
Jeong, Minbyul and
Lim, Suhyeon and
Jung, Sungwook and
Yun, Jaehoon and
Roh, Taeyun and
Sung, Mujeen and
Kang, Jaewoo",
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.1950/",
pages = "44337--44375",
ISBN = "979-8-89176-459-0",
abstract = "Recent multimodal large language models have shown great promise in clinical image reasoning, but existing post-training pipelines remain predominantly outcome-centric, relying on final answer correctness or sequence-level preferences. This approach suffers from sparse credit assignment, making it difficult to optimize the internal reasoning process essential for clinical applications. Our analysis reveals that cascading errors from early-stage reasoning failures are a leading cause of incorrect predictions in medical visual question answering (VQA) benchmarks. Motivated by this, we propose Medical Reasoning-aware Policy Optimization(MRPO), an RL algorithm that incorporates step-wise process rewards. When the final answer is incorrect, MRPO assigns exponentially larger penalties to tokens in earlier invalid reasoning steps, breaking failure cascades without compromising successful paths. Across three multimodal LLM backbones, MRPO consistently outperforms standard GRPO and a recent RL baseline, and on Qwen3-VL-8B-Instruct even outperforms substantially larger medical MLLMs such as HuatuoGPT-Vision-34B by 2.79 points. Moreover, MRPO reduces early-stage reasoning failures from 64.0{\%} to 13.0{\%}, demonstrating that targeted mitigation of cascading failures improves both reasoning quality and final answer accuracy."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="jung-etal-2026-breaking">
<titleInfo>
<title>Breaking Failure Cascades: Step-Aware Reinforcement Learning for Medical Multimodal Reasoning</title>
</titleInfo>
<name type="personal">
<namePart type="given">Junha</namePart>
<namePart type="family">Jung</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Minbyul</namePart>
<namePart type="family">Jeong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Suhyeon</namePart>
<namePart type="family">Lim</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sungwook</namePart>
<namePart type="family">Jung</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jaehoon</namePart>
<namePart type="family">Yun</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Taeyun</namePart>
<namePart type="family">Roh</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mujeen</namePart>
<namePart type="family">Sung</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jaewoo</namePart>
<namePart type="family">Kang</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>Recent multimodal large language models have shown great promise in clinical image reasoning, but existing post-training pipelines remain predominantly outcome-centric, relying on final answer correctness or sequence-level preferences. This approach suffers from sparse credit assignment, making it difficult to optimize the internal reasoning process essential for clinical applications. Our analysis reveals that cascading errors from early-stage reasoning failures are a leading cause of incorrect predictions in medical visual question answering (VQA) benchmarks. Motivated by this, we propose Medical Reasoning-aware Policy Optimization(MRPO), an RL algorithm that incorporates step-wise process rewards. When the final answer is incorrect, MRPO assigns exponentially larger penalties to tokens in earlier invalid reasoning steps, breaking failure cascades without compromising successful paths. Across three multimodal LLM backbones, MRPO consistently outperforms standard GRPO and a recent RL baseline, and on Qwen3-VL-8B-Instruct even outperforms substantially larger medical MLLMs such as HuatuoGPT-Vision-34B by 2.79 points. Moreover, MRPO reduces early-stage reasoning failures from 64.0% to 13.0%, demonstrating that targeted mitigation of cascading failures improves both reasoning quality and final answer accuracy.</abstract>
<identifier type="citekey">jung-etal-2026-breaking</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.1950/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>44337</start>
<end>44375</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Breaking Failure Cascades: Step-Aware Reinforcement Learning for Medical Multimodal Reasoning
%A Jung, Junha
%A Jeong, Minbyul
%A Lim, Suhyeon
%A Jung, Sungwook
%A Yun, Jaehoon
%A Roh, Taeyun
%A Sung, Mujeen
%A Kang, Jaewoo
%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 jung-etal-2026-breaking
%X Recent multimodal large language models have shown great promise in clinical image reasoning, but existing post-training pipelines remain predominantly outcome-centric, relying on final answer correctness or sequence-level preferences. This approach suffers from sparse credit assignment, making it difficult to optimize the internal reasoning process essential for clinical applications. Our analysis reveals that cascading errors from early-stage reasoning failures are a leading cause of incorrect predictions in medical visual question answering (VQA) benchmarks. Motivated by this, we propose Medical Reasoning-aware Policy Optimization(MRPO), an RL algorithm that incorporates step-wise process rewards. When the final answer is incorrect, MRPO assigns exponentially larger penalties to tokens in earlier invalid reasoning steps, breaking failure cascades without compromising successful paths. Across three multimodal LLM backbones, MRPO consistently outperforms standard GRPO and a recent RL baseline, and on Qwen3-VL-8B-Instruct even outperforms substantially larger medical MLLMs such as HuatuoGPT-Vision-34B by 2.79 points. Moreover, MRPO reduces early-stage reasoning failures from 64.0% to 13.0%, demonstrating that targeted mitigation of cascading failures improves both reasoning quality and final answer accuracy.
%U https://aclanthology.org/2026.emnlp-main.1950/
%P 44337-44375
Markdown (Informal)
[Breaking Failure Cascades: Step-Aware Reinforcement Learning for Medical Multimodal Reasoning](https://aclanthology.org/2026.emnlp-main.1950/) (Jung et al., EMNLP 2026)
ACL
- Junha Jung, Minbyul Jeong, Suhyeon Lim, Sungwook Jung, Jaehoon Yun, Taeyun Roh, Mujeen Sung, and Jaewoo Kang. 2026. Breaking Failure Cascades: Step-Aware Reinforcement Learning for Medical Multimodal Reasoning. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 44337–44375, Budapest, Hungary. Association for Computational Linguistics.