Table 7 Transferability of adversarial samples when both the source model (with Mixed LSTM architecture) and target models (with other architectures) are BEAM-related models. The N of GPBEAM-DE is 5. Since the adversarial samples are all obtained from the source model, the values of \({DL}_{B}\) or \({DL}_{E}\) are the same for all architectures, and so only the \({DL}_{B}\) and \({DL}_{E}\) of Mixed LSTM (source) are shown here

From: Perturbing BEAMs: EEG adversarial attack to deep learning models for epilepsy diagnosing