资讯

An autoencoder is a neural network that predicts its own input. The diagram in Figure 3 shows the architecture of the 65-32-8-32-65 autoencoder used in the demo program.
An autoencoder is a neural network that predicts its own input. The diagram in Figure 3 shows the architecture of the 65-32-8-32-65 autoencoder used in the demo program.
We propose an unsupervised method for detecting adversarial attacks in inner layers of autoencoder (AE) networks by maximizing a non-parametric measure of anomalous node activations.