Skip to main content
. 2020 May 4;22(2):1767–1781. doi: 10.1093/bib/bbaa021

Figure 1.

Figure 1

DL architectures used in BA estimation (A) dense (or deep) neural network (DNN), (B) long short-term memory (LSTM) cells (LSTM updates for timestep Inline graphic given the input Inline graphic and the previous state Inline graphic and previous cell output Inline graphic), (C) convolutional neural network (CNN) and (D) convoutional LSTM cell. Here (A) and (C) show the network while (B) and (D) show the basic structure; to use the models for age estimation, we add dense layers and a single unit output.