Fig. 14

The one-class classification training procedure: the encoder and decoder parts of the auto-encoder model shown in Fig. 13 are pre-trained and fixed (as indicated by a “*”); the OC-SVM is trained on the outputs of \(\mathcal {D}_{{\textrm{t}} \hat{\textrm{t}}}\) and \(\mathcal {D}_{\textrm{t}}\) terms that are the results of \(I^L_{\varvec{\phi }}({\textbf {X}};{\textbf {T}})\) decomposition and the \(\mathcal {D}_{{\textrm{x}} \hat{\textrm{x}}}\) and \(\mathcal {D}_{\textrm{x}}\) terms that are the results of \(I^L_{\varvec{\phi },\varvec{\theta }}({\textbf {T}};{\textbf {X}})\) decomposition