Algorithm 1 Update the trainable parameter $\mathit{\theta }$ of WPNF via the stochastic gradient descent method of momentum

Require: learning rate $\eta$, Momentum parameter $\alpha$, Initial parameters $\theta$, Initial speed $v$

For epoch to set value, do:

For episode to set value, do:

Randomly take $m$ samples $\left\{x^{\left(1\right)},x^{\left(2\right)},x^{\left(3\right)}\dots x^{\left(m\right)}\right\}$ from the query set, and their true labels are $c_k{}^{\left(i\right)}$

Compute gradient of the samples: $g\leftarrow \frac{1}{m}{\nabla }_{\theta }{\displaystyle \sum }_{i}J\left(x, c_k\right)$

Compute speed update: $v\leftarrow \alpha a-\eta g$

Update parameters: $\theta \leftarrow \theta +v$

End For

End For