Algorithm 1. TRAINING INTERATION PROCEDURE |
|
Require: is the batch size, is a batch of data with size , and donates each sample in a batch. is score loss, is boundary loss, is a hyperparameter, denotes the model parameters, denotes the updated model parameters. is the learning rate. |
1:
|
2: for
|
3:
|
4:
|
5:
|
6: end for
|
7:
|
8: return
|