Fig. 3
From: Automatic identification of differences in behavioral co-occurrence between groups
Feature learning using an autoencoder. An autoencoder is a neural network that learns the latent representation of an input. The encoder converts the input into the latent representation, the hidden layer contains the latent representation of the input data, and the decoder reconstructs the original input from the latent representation. Each circle is regarded as a unit in the neural network. The number of units in the input layer and that of the output layer are identical to the dimension of the input data. The number of units in the hidden layer is less than that in the input layer. By limiting the number of hidden units, we can compress the input data. Because the compressed representation contains important information necessary to reconstruct the output, the compressed data are considered to represent latent states of an animal (i.e., behavioral modes)