Generalizable Sample-Efficient Siamese Autoencoder for Tinnitus Diagnosis in Listeners With Subjective Tinnitus

Electroencephalogram (EEG)-based neurofeedback has been widely studied for tinnitus therapy in recent years. Most existing research relies on experts' cognitive prediction, and studies based machine learning deep are either data-hungry or not well generalizable to new subjects. In this paper, we propose a robust, data-efficient model distinguishing from the healthy state EEG-based neurofeedback. We trend descriptor, feature extractor with lower fineness, reduce effect of electrode noises EEG signals, siamese encoder-decoder network boosted supervised manner learn accurate alignment acquire high-quality transferable mappings across subjects signal channels. Our experiments show proposed method significantly outperforms state-of-the-art algorithms when analyzing subjects' 90dB 100dB sound, achieving an accuracy 91.67%-94.44% predicting control subject-independent setting. ablation mixed parameters method's stability performance.