Our work, ProDiF, has been accepted to ICLR Workshop 2025! It provides comprehensive protection for on-device ML models against model extraction and subsequent unauthorized fine-tuning.