Live Demonstration: Real-time EEG-based Affective Computing Using On-chip Learning Long-term Recurrent Convolutional Network

An edge artificial intelligence (AI) affective computing system based on electroencephalogram (EEG) will be demonstrated for multi-class emotional classification. It's composed of a dry electrode EEG headset, RISC-V feature extraction processor, long-term recurrent convolutional network (LRCN) on-chip platform, and graphical user interface (GUI). The LRCN platform is implemented with a TSMC 16-nm FinFET technology chip for efficient edge AI application included training and acceleration. Bluetooth 2.1 modules are deployed to construct a complete wireless edge-AI system from front-end to back-end. It takes 350 ms to identify and demonstrate one emotion state from the EEG headset front-end to the GUI display back-end.