ConfProBench: A Confidence Evaluation Benchmark for MLLM-Based Process Judges
Fuente:
arXiv
Salvato in:
| Autori principali: | , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866913977762578432 |
|---|---|
| author | Zhou, Yue Chang, Yi Wu, Yuan |
| author_facet | Zhou, Yue Chang, Yi Wu, Yuan |
| contents | Reasoning is a critical capability of multimodal large language models (MLLMs) for solving complex multimodal tasks, and judging the correctness of reasoning steps is crucial for improving this capability. Recently, MLLM-based process judges (MPJs) have been widely used to assess the correctness of reasoning steps in multimodal tasks. Therefore, evaluating MPJs is important for identifying their limitations and guiding future improvements. However, existing benchmarks for MPJs mainly focus on tasks such as step correctness classification and reasoning process search, while overlooking a key aspect: whether the confidence scores produced by MPJs at the step level are reliable. To address this gap, we propose ConfProBench, the first comprehensive benchmark designed to systematically evaluate the reliability of step-level confidence scores generated by MPJs. Our benchmark constructs three types of adversarially perturbed reasoning steps: Synonym Substitution, Syntactic Transformation, and Image Perturbation, to test the robustness of MPJ confidence under perturbations. In addition, we introduce three novel evaluation metrics: Confidence Robustness Score (CRS), Confidence Sensitivity Score (CSS), and Confidence Calibration Score (CCS), which evaluate robustness, sensitivity, and calibration, respectively. We evaluate 14 state-of-the-art MLLMs, including both proprietary and open-source models. Experiments reveal limitations in current MPJs' confidence performance and offer competitive baselines to support future research. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2508_04576 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | ConfProBench: A Confidence Evaluation Benchmark for MLLM-Based Process Judges Zhou, Yue Chang, Yi Wu, Yuan Artificial Intelligence I.2.6; I.2.7; D.2.8 Reasoning is a critical capability of multimodal large language models (MLLMs) for solving complex multimodal tasks, and judging the correctness of reasoning steps is crucial for improving this capability. Recently, MLLM-based process judges (MPJs) have been widely used to assess the correctness of reasoning steps in multimodal tasks. Therefore, evaluating MPJs is important for identifying their limitations and guiding future improvements. However, existing benchmarks for MPJs mainly focus on tasks such as step correctness classification and reasoning process search, while overlooking a key aspect: whether the confidence scores produced by MPJs at the step level are reliable. To address this gap, we propose ConfProBench, the first comprehensive benchmark designed to systematically evaluate the reliability of step-level confidence scores generated by MPJs. Our benchmark constructs three types of adversarially perturbed reasoning steps: Synonym Substitution, Syntactic Transformation, and Image Perturbation, to test the robustness of MPJ confidence under perturbations. In addition, we introduce three novel evaluation metrics: Confidence Robustness Score (CRS), Confidence Sensitivity Score (CSS), and Confidence Calibration Score (CCS), which evaluate robustness, sensitivity, and calibration, respectively. We evaluate 14 state-of-the-art MLLMs, including both proprietary and open-source models. Experiments reveal limitations in current MPJs' confidence performance and offer competitive baselines to support future research. |
| title | ConfProBench: A Confidence Evaluation Benchmark for MLLM-Based Process Judges |
| topic | Artificial Intelligence I.2.6; I.2.7; D.2.8 |
| url | https://arxiv.org/abs/2508.04576 |