Deep Neural Network Watermarking

Deep neural network (DNN) watermarking aims to embed imperceptible yet robust identifiers into DNN models to protect intellectual property. Current research focuses on improving watermark robustness against various attacks, including geometric distortions and adversarial manipulations, often employing transformer architectures like Swin Transformers alongside convolutional neural networks to achieve this. This field is crucial for securing the ownership of increasingly valuable DNN models and is actively developing new techniques to address vulnerabilities and improve the reliability of watermarking schemes in both centralized and federated learning environments.