Attention-Based Sequence-to-Sequence Model for Time Series Imputation

. 2022 Dec 9;24(12):1798. doi: 10.3390/e24121798

Algorithm 1 Multi-Head Attention Algorithm.

Input:

T_{l}

T_{r}

of the time series, n the number of attention
Output: U the result of multi-head attention
for h = 0 to n do
I←

f_{Φ} (T_{l}, T_{r})

P←

φ_{Θ} (T_{l}, T_{r})

Q, K, V

←

L_{Z} (I + P)

W←

S (\frac{Q K^{T}}{\sqrt{d_{k}}})

A_{i}

←

W \cdot V

U←

L_{Z} (A_{1} \cup A_{2} \cup \dots, A_{n})

end for
return U