Generalized hidden-mapping transductive transfer learning for recognition of epileptic electroencephalogram signals

Abstract

Electroencephalogram (EEG) signal identification based on intelligent models is an important means in epilepsy detection. In the recognition of epileptic EEG signals, traditional intelligent methods usually assume that the training dataset and testing dataset have the same distribution, and the data available for training are adequate. However, these two conditions cannot always be met in practice, which reduces the ability of the intelligent recognition model obtained in detecting epileptic EEG signals. To overcome this issue, an effective strategy is to introduce transfer learning in the construction of the intelligent models, where knowledge is learned from the related scenes (source domains) to enhance the performance of model trained in the current scene (target domain). Although transfer learning has been used in EEG signal identification, many existing transfer learning techniques are designed only for a specific intelligent model, which limit their applicability to other classical intelligent models. To extend the scope of application, the generalized hidden-mapping transductive learning method is proposed to realize transfer learning for several classical intelligent models, including feedforward neural networks, fuzzy systems, and kernelized linear models. These intelligent models can be trained effectively by the proposed method even though the data available are insufficient for model training, and the generalization abilities of the trained model is also enhanced by transductive learning. A number of experiments are carried out to demonstrate the effectiveness of the proposed method in epileptic EEG recognition. The results show that the method is highly competitive or superior to some existing state-of-the-art methods.