Abstract: In epileptic electroencephalogram (EEG) signal recognition, traditional intelligent modeling method requires that the training data set and test data sets are subject to the same distribution. But in practical applications, some cases cannot meet this condition, which results in a sharp degeneration in the performance of traditional methods. In order to overcome the above challenge, by the introduction of transfer learning strategies, this paper proposes a transductive radial basis function neural network (TRBFNN) in order to cater for the migration environment of data distribution. The experimental results in epileptic EEG signals recognition show that the proposed TRBFNN has better adaptability for migration scenarios of data distribution and the recognition performance will not be sharp deterioration when there are differences between the distributions of the training data and test data.

Key words: electroencephalogram (EEG), radial basis function neural network, transductive transfer learning

摘要： 在癫痫脑电图（electroencephalogram，EEG）信号识别中，传统的智能建模方法要求训练数据集和测试数据集均服从相同的分布。但在实际应用中，某些情况并不能满足此条件，进而导致传统方法性能急剧下降。针对上述情况，引入迁移学习策略，提出了适用于数据分布迁移环境的直推式径向基神经网络（transductive radial basis function neural network，TRBFNN）。该方法在癫痫EEG信号识别中的实验结果表明：直推式径向基神经网络具有较好的场景迁移适应性，对训练数据和测试数据存在差异时，识别性能不会出现急剧恶化的现象。

关键词: 脑电图（EEG）, 径向基神经网络, 直推式迁移学习