The study aims to achieve a high information transfer rate (ITR) by using a deep neural network (DNN) that is trained using data from source domains using target data that is unlabeled. The research was performed using an adaptation that minimized loss using self-adaptation through pseudo-labels and a regularity term to obtain data structure, getting rid of the necessity for calibration. In the findings, it is observed that the studied method could achieve very high ITRs or 201.15 bits/min on the benchmark dataset and 145.02 bits/min on the BETA dataset, which is higher than state-of-the-art methods. The studied method uses a local-regularity term to ensure that the data is labeled consistently, which improves the adaptation reliability to the SSVEP signals of new users. The method allows for use of SSVEP-based BCI spellers without the need for recalibration while maintaining high performance accuracy.