Unveiling Deep Learning Models: Leveraging Active Learning and High-Confidence Feature Extraction in Model Stealing