… the details of model parameters and therefore should be obfuscated with priority. To this
end, we propose a boundary differential private layer (BDPL) for machine learning services. …

… propose boundary differential privacy (BDP) against such attacks … a boundary differentially
private layer (BDPL) for machine … against model extraction attacks with respect to -boundary …

… -based work proposed boundary differential private layer for defending model extraction attempt
with … Zheng et al. proposed a boundary differentially private layer (BDPL) [13] to withhold …

… -down order of attack surfaces, from model prediction to data … the extraction of private decision
boundary on machine learning … a boundary differentially private layer (BDPL) for machine …

… machine learning models and steal the deployed model … boundary differentially private layer
(BDPL) for binary classifier. It perturbs the model outputs by DP to mitigate model extraction, …

… who aims to conduct model extraction attack on a cloud-based model which is protected by
… BDPL uses differential privacy to perturb the output [12]. A boundary differential privacy layer …

… a differentially private SGD algorithm in deep learning, and … learn the decision boundary
and the model extraction attack is … that BDPL is only suitable for the logistic regression model, …

… of the \(\epsilon \)-differentially private mechanisms, which is … methods against model
extraction attacks, ie, model output … algorithm defined within our boundary differential privacy …

… the model extraction attack on machine learning models (eg … the harm of the model extraction
attack in the experiments. … [16] propose boundary differentially private layer (BDPL) for …

… Machine learning models are proprietary in that they encompass … of attack-defense. One
potential candidate for the defense mechanism is the boundary differentially private layer (BDPL)…