Few Shot Domain Adaptation Using Transformer and GNN-Based Fine Tuning

This research paper introduces a novel cross-domain few-shot learning framework that combines a Domain Adaptation Transformer (DAT) with a Graph Neural Network (GNN)-based Fine Tuning model. The framework aims to enhance model adaptation across diverse domains. The proposed approach achieves notable accuracy improvements on two distinct datasets, the Office dataset and the Crop Disease dataset. The DAT model achieves 75% accuracy on the Office dataset, while the GNN-based Fine Tuning model attains a remarkable 96.97% accuracy on the Crop Disease dataset—outperforming the existing benchmark of 95.28%. These results demonstrate the effectiveness of the proposed framework in achieving higher accuracy across various domains.