Algorithm 1. Transformer-Enhanced Autoencoder Construction

Input: X

Output: Branch A: Encoder (E); $\widehat{ X}, L_{AE}$

1: procedure Transformer-enhanced Autoencoder

2: Input Layer ← Input

3: The intermediate feature A is calculated according to the multi-head attention mechanism

4: Layernormalization Layer yields intermediate variables p

5: Compute Encoder output $z$ according to Equation (6)

6: Compute the reconstruction matrix $\widehat{X}$ according to Equation (7)

7: Calculate the Latent feature matrix E according to Equation (8)

8: Compute the reconstruction loss $L_{AE}$ according to Equations (9)–(11)

9: end procedure