The paper proposes a method for intellectual property protection for a learned deep model. The problem is highly relevant and understudied. While the proposed method was considered too incremental with added complexity compared to previous work, there was enough new perspectives and ideas to warrant acceptance.