Table 13 Ablation study results: Impact of key architectural components.

From: Secure multi-party test case data generation through generative adversarial networks

Configuration

Privacy (GLR \(\downarrow\))

Efficiency (FRT \(\downarrow\))

Syntax (SCR \(\uparrow\))

Diversity (ATD\(\uparrow\))

FAT-CG (Full)

9.7%

185 ms

93.8%

8.7

w/o Autoencoder

9.5%

320 ms

94.0%

8.8

w/o Paillier HE

78.2%

110 ms

93.5%

8.6

w/o Adversarial

9.2%

180 ms

94.1%

3.5

  1. \(^{-}\) “w/o Syntax Verifier” denotes training without the syntax regularization term (\(\lambda _{syn} = 0\)), leading to a failure in generating valid protocol messages.
Back to article page