Published Paper Details:

Abstract--Multimodal learning empowers large language models (LLMs) to process text, images, and videos, enabling advanced capabilities in tasks like visual question answering (VQA), image captioning, and video understanding. This paper introduces the Multimodal Fusion Transformer (MFT), a novel architecture that employs a hybrid attention mechanism to dynamically align and fuse features from heterogeneous modalities. The MFT achieves 10-15% performance improvements over state-of-the-art models on benchmarks such as VQA v2.0, COCO, and MSRVTT, while reducing computational overhead. A comprehensive literature review identifies gaps in existing approaches, and a compact flow diagram illustrates the MFT architecture. Extensive experiments, including ablation studies and robustness analysis, validate its efficiency, scalability, and adaptability for real-world multimodal tasks. Index Terms--Multimodal Learning, Large Language Models, Text-Image-Video Integration, Hybrid Attention, Transformer Architecture