Figure 2
Left (a) Training parameters for the VAE of the proposed architecture. The dimension of \(\varvec{z}\) is the output number of the encoder. The training size rate is the ratio of the total number of data to the data size of the input at training. Regarding the architecture evaluation, the input size is set to \((1 - \text {Training size rate})\). The learning rate is the initial learning rate, and the optimiser used is Adam31. Right (a) Training parameters for the IIC of the proposed architecture. The number of output classes is set to the number of classes to be classified. The classifier number is for multiple classifiers that are used to improve the performance of the classifier using spectral clustering. (b) Training curve during the training and evaluation of the VAE. The solid and dashed lines in the figure show the training objective \(\delta \equiv -\sum _{i}^ {N}\mathcal {L}(\varvec{x}^{(i)}, \varvec{\theta }, \varvec{\phi })\) at the time of training and evaluation, respectively. (c) Reconstructed images generated by the decoder of the VAE at 100 epochs in Case 3.
