Salvato in:
Dettagli Bibliografici
Autori principali: Hu, Pengfei, Zhang, Zhenrong, Chang, Qikai, Liu, Shuhang, Ma, Jiefeng, Du, Jun, Zhang, Jianshu, Liu, Quan, Gao, Jianqing, Ma, Feng, Liu, Qingfeng
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:https://arxiv.org/abs/2504.10222
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866910910887493632
author Hu, Pengfei
Zhang, Zhenrong
Chang, Qikai
Liu, Shuhang
Ma, Jiefeng
Du, Jun
Zhang, Jianshu
Liu, Quan
Gao, Jianqing
Ma, Feng
Liu, Qingfeng
author_facet Hu, Pengfei
Zhang, Zhenrong
Chang, Qikai
Liu, Shuhang
Ma, Jiefeng
Du, Jun
Zhang, Jianshu
Liu, Quan
Gao, Jianqing
Ma, Feng
Liu, Qingfeng
contents Recent work increasingly focuses on improving the reasoning capabilities of Multimodal Large Language Models (MLLMs). Among existing methods, Process Reward Models (PRMs) stand out for offering dense, step-wise supervision to guide intermediate reasoning. However, how to effectively integrate PRMs into search strategies remains an open question. In this paper, we introduce PRM-BAS (PRM-Guided Beam Annealing Search), a lightweight approach for PRM-guided reasoning that dynamically adjusts beam size -- starting with a broader search space and gradually narrowing it as contextual information accumulates, thereby balancing performance and efficiency. We further propose a unified framework for data construction and PRM training. Specifically, we construct the PRM-BAS-300k dataset by selecting 300k questions from existing datasets and performing rollouts at each step to estimate the probability of reaching a correct final answer. The PRM is then trained using a combination of value loss for absolute action quality and rank loss for relative action quality. Extensive experiments on challenging multimodal reasoning benchmarks demonstrate that PRM-BAS significantly improves reasoning performance while maintaining low computational cost. Moreover, it generalizes well across different model scales and architectures, showcasing strong robustness and plug-and-play capability.
format Preprint
id arxiv_https___arxiv_org_abs_2504_10222
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle PRM-BAS: Enhancing Multimodal Reasoning through PRM-guided Beam Annealing Search
Hu, Pengfei
Zhang, Zhenrong
Chang, Qikai
Liu, Shuhang
Ma, Jiefeng
Du, Jun
Zhang, Jianshu
Liu, Quan
Gao, Jianqing
Ma, Feng
Liu, Qingfeng
Multimedia
Recent work increasingly focuses on improving the reasoning capabilities of Multimodal Large Language Models (MLLMs). Among existing methods, Process Reward Models (PRMs) stand out for offering dense, step-wise supervision to guide intermediate reasoning. However, how to effectively integrate PRMs into search strategies remains an open question. In this paper, we introduce PRM-BAS (PRM-Guided Beam Annealing Search), a lightweight approach for PRM-guided reasoning that dynamically adjusts beam size -- starting with a broader search space and gradually narrowing it as contextual information accumulates, thereby balancing performance and efficiency. We further propose a unified framework for data construction and PRM training. Specifically, we construct the PRM-BAS-300k dataset by selecting 300k questions from existing datasets and performing rollouts at each step to estimate the probability of reaching a correct final answer. The PRM is then trained using a combination of value loss for absolute action quality and rank loss for relative action quality. Extensive experiments on challenging multimodal reasoning benchmarks demonstrate that PRM-BAS significantly improves reasoning performance while maintaining low computational cost. Moreover, it generalizes well across different model scales and architectures, showcasing strong robustness and plug-and-play capability.
title PRM-BAS: Enhancing Multimodal Reasoning through PRM-guided Beam Annealing Search
topic Multimedia
url https://arxiv.org/abs/2504.10222