Deep learning has achieved remarkable progress in various applications heightening the
importance of safeguarding the intellectual property (IP) of well-trained models. It entails not
only authorizing usage but also ensuring the deployment of models in authorized data
domains ie making models exclusive to certain target domains. Previous methods
necessitate concurrent access to source training data and target unauthorized data when
performing IP protection making them risky and inefficient for decentralized private data. In …

Lemma 1 ([8]). Let p be the predicted label outputted by a representation model when
feeding with input x, and suppose that p is a scalar random variable and x is balanced on
the ground truth label y. And P (·) is the distribution. If the KL divergence loss KL (P (p)∥ P
(y)) increases, the mutual information I (z; y) will decrease.