Evaluated CVAE architectures. Each architecture was trained with different randomized model weight initializations across 25 instances to compute the mean MSE reconstruction error.

Evaluated CVAE architectures. Each architecture was trained with different randomized model weight initializations across 25 instances to compute the mean MSE reconstruction error.本研究评估了条件变分自编码器（Conditional Variational Autoencoder，CVAE）的多种架构。针对每种架构，均采用25组不同的随机模型权重初始化方式开展训练，以此计算均方重建误差（Mean Squared Error，MSE）的均值。