Fig. 4

Framework of HTM for anomaly detection [34, Fig 3(a)] Input vector \(\vec {x}_{t}\) of the data stream is sent to the HTM where it produces a sparse binary vector representing the current input, \(a(\vec {x}_{t})\) and prediction for the next time step, \(\vec {\pi }(\vec {x}_{t})\). The prediction error, \(s_t\) is calculated and is used to obtain the anomaly likelihood, \(L_t\). Image obtained under the Creative Commons license. No revisions were made to the image.