This is an online seminar. Registration is required.
【 Computational brain dynamics team】
*【Speaker】Li Yuanhao, Tokyo Institute of Technology *
Title: Robust Brain Activity Decoding Based on Information Theoretic Learning
Brain activity decoding aims to predict the intention or mental status of the brain by the utilization of brain signals. To realize feature extraction and pattern recognition for the pre-processed brain recordings, machine learning techniques are widely utilized in the brain activity decoding tasks. However, existing methods of brain activity recording, such as electroencephalogram (EEG) or functional magnetic resonance imaging (fMRI), are prone to the interference of physiological artifacts and other environmental noises, which are usually intractable for the conventional algorithms, leading to unsatisfactory brain decoding performance. To address this issue, my research focuses on proposing new robust brain decoding algorithms under the framework of information theoretic learning, so as to reduce the negative effects of the brain recording noise. In particular, minimum error entropy criterion and maximum correntropy criterion were utilized to structure new robust objective functions for brain decoding algorithms. The proposed algorithms were evaluated systematically with synthetic datasets and real-world brain datasets. Experimental results demonstrated that the information-theoretic-learning-based methods can effectively alleviate the performance deterioration caused by the noises and realize higher brain decoding accuracy on the real-world noisy brain data, thus promoting the investigation of brain mechanism and the development of brain-computer interface.
Public events of RIKEN Center for Advanced Intelligence Project (AIP)Join community