What is Zero-Shot Learning, and how does it enable models to make predictions on unseen data? How does Zero-Shot Learning differ from traditional supervised learning approaches? What role do pre-trained models and semantic understanding play in Zero-Shot Learning? What are the advantages and limitations of using Zero-Shot Learning in real-world applications? How is Zero-Shot Learning applied in fields such as natural language processing and computer vision?