LONG SHORT-TERM MEMORY MODEL WITH THE EXTERNAL TREND AND INTERNAL COMPONENTS ANALYSIS

This paper presents a modification of a recurrent neural network with long-term and short-term memory for modeling electroencephalogram signals and highlights its potential in predicting pathological conditions. The demonstrated interpretation includes a method for decomposing the external trend and internal components that most characteristically determine the parameters of the input signal. The obtained segmented data, after pre-processing by this method, are further processed by a neural network, which is trained on them with more efficient calibration. As a result, the trained network can reproduce the data on which it was trained and predict the further trend of brain activity. The results show the potential of this approach for both basic neuroscience research and clinical applications in the diagnosis and modeling of neuropathologies.