Fig. 7
From: Biomedical semantic indexing by deep neural network with multi-task learning
The classification loss during the training process for different Model. The result shows that network with SMTL converges faster and has better performance than networks with an ordinary parallel multi-task learning structure (Model A) and network without a multi-task structure (Model B). It also shows that BGRU has better performance than unidirectional GRU (Model C) in SMTL