TADW: Traceable and Anti-detection Dynamic Watermarking of Deep Neural Networks

Figure 3

The testing accuracy of the watermarked model under different pruning rates (SST-2).