EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Prompt-Driven Fuzzing Debiasing Framework for Robust Visual Question Answering

Yali Fan, Gangyu Huang, Qiwen Lu, Shengbo Chen

Multimodal Technologies and Interaction · 31.07.2026 · 10 (8) · 82

DOI: 10.3390/mti10080082 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Multimodal Technologies and Interaction, online erschienen am 31.07.2026. Verfasst von Yali Fan und 3 weiteren Autor:innen. Die Arbeit nennt 59 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Natural Science Foundation of Henan Province, High level Overseas Talent Project of Jiangxi Province, 2024 Jiangxi Ganpo Talent Plan.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Visual Question Answering (VQA) systems have achieved impressive performance with the rise of large-scale vision–language models (VLMs). However, these models remain vulnerable to multiple forms of multimodal bias, severely limiting their robustness and generalization. Existing debiasing techniques mainly depend on post hoc evaluation or architectural modifications, while recent prompt-learning-based methods reveal new opportunities for aligning downstream tasks with pretrained models. In this work, we propose a unified prompt-driven debiasing framework that integrates generative prompt learning and a fuzzing-based bias correction mechanism. The generative prompt component reformulates VQA as a cloze-style masked prediction problem, leveraging pretrained language priors to improve semantic grounding. Meanwhile, the fuzzing-based module actively constructs unexpected test samples during training and employs a reflection mechanism to correct biased predictions in-loop, yielding inference-time robustness without additional test-time components. Extensive experiments on VQA-v2, VQA-CP, VQA-CE, GQA-OOD, and VQA-VS demonstrate that the proposed framework significantly improves both in-distribution (ID) accuracy and out-of-distribution (OOD) robustness, outperforming existing prompt-only or data-augmentation-only debiasing methods.

Förderung

  • Natural Science Foundation of Henan Province
  • High level Overseas Talent Project of Jiangxi Province
  • 2024 Jiangxi Ganpo Talent Plan
  • Ji’ An “Revealed Challenge” Key Core Common Technology List Project

Zitieren

APA

Fan, Y., Huang, G., Lu, Q., & Chen, S. (2026). Prompt-Driven Fuzzing Debiasing Framework for Robust Visual Question Answering. Multimodal Technologies and Interaction, 10(8), 82. https://doi.org/10.3390/mti10080082

Vancouver

Fan Y, Huang G, Lu Q, Chen S. Prompt-Driven Fuzzing Debiasing Framework for Robust Visual Question Answering. Multimodal Technologies and Interaction. 2026;10(8):82. doi:10.3390/mti10080082

BibTeX
@article{Fan2026,
  author = {Fan, Yali and Huang, Gangyu and Lu, Qiwen and Chen, Shengbo},
  title = {Prompt-Driven Fuzzing Debiasing Framework for Robust Visual Question Answering},
  journal = {Multimodal Technologies and Interaction},
  year = {2026},
  volume = {10},
  number = {8},
  pages = {82},
  doi = {10.3390/mti10080082}
}
Kostenlos starten →