Backdoor Defense for Data-Free Distillation with Poisoned Teachers

The task of Backdoor Defense for Data-Free Distillation with Poisoned Teachers aims to ensure that the student model does not inherit malicious behaviors through detection and mitigation techniques in the process of data-free distillation, where the teacher model is compromised. The goal of this task is to enhance the security and robustness of the model, prevent potential privacy leaks and security threats, thereby safeguarding the credibility and reliability of machine learning systems.