by Pratheswaran Hariharan, Haiping Xu and Donghui Yan
Journal of Engineering Research and Sciences, Volume 5, Issue 9, Page # 1-14, 2026; DOI: 10.55708/js0509001
Abstract: Multimodal large language models (MLLMs) can generate fluent visual responses even when the underlying visual prediction is weak, ambiguous, or incorrect. This work presents a retrieval-augmented, reliability-aware selective inference method that evaluates the strength and consistency of visual evidence before a prediction is communicated through a downstream multimodal response. A pretrained ResNet-50 encoder extracts normalized… Read More

(This article belongs to the Special Issue on SP8 (Special Issue on Digital and Engineering Transformations in Science and Technology (SI-DETST-26)))