Table 25 Summary of model limitations and mitigation strategies.
Limitation | Description | Proposed Mitigation |
|---|---|---|
Computational Resources | Requires high-performance GPUs with significant memory (16 GB +) for training and inference | Employ model compression techniques or explore distributed training to optimize hardware utilization |
Training Time | Longer training times due to the complexity of the ensemble model integrating multiple components | Optimize training pipelines and consider reducing the number of ensemble components without sacrificing accuracy |
Overfitting | Higher risk of overfitting, especially on smaller datasets, due to the complexity of the model | Utilize regularization techniques (e.g., dropout, L2), extensive data augmentation, and cross-validation |
Interpretability | Reduced interpretability compared to simpler models, even with techniques like Grad-CAM | Extend interpretability methods to cover all components, such as visualization tools for ViT and XGBoost |
Scalability | Limited scalability to resource-constrained environments like mobile or edge devices | Develop lightweight versions of the ensemble using knowledge distillation or pruning methods |
