Algorithm 1: FIS using top-k strategy.

Input: Dictionary of orders: $\mathbb{D}$. Output: Hard vector of gated unit: $\mathbb{H}$. $\mathbb{W}\leftarrow$ Randomly initialize weights of features; $\mathbb{S}\leftarrow$ Compute Gumbel-softmax of $\mathbb{W}$; $k\leftarrow$ Get order of unit u from $\mathbb{D}$; $inde{x}^{\left(k\right)}\leftarrow$ indexes of the top-k probabilities in $\mathbb{S}$; $\mathbb{H}\leftarrow$ Generate multi-hot vector based on $inde{x}^{\left(k\right)}$.