Toward Generalizable Representations and Scalable Citizen Science for Brain-Computer Interfaces

This talk will explore how large-scale self-supervised learning, combined with our latest multimodal neurotechnology—the newly released Muse integrating EEG and functional near-infrared spectroscopy (fNIRS)—and an open citizen science platform, accelerates the development of robust and adaptive neurotechnologies. We will present the technological capabilities of Muse headbands, which provide research-grade EEG, PPG, fNIRS, and inertial measurements in a comfortable, accessible, and consumer-friendly form factor. We will discuss how widespread consumer adoption has facilitated extensive neural data collection during everyday brain health-oriented activities beyond traditional laboratory environments. Leveraging our substantial global user base through a citizen science platform where researchers world-wide can now conduct large-scale studies, including adaptive closed-loop neural applications, event-related potential (ERP) measurements, and comprehensive behavioral and self-report data collection. From an AI perspective, this talk will highlight how these expansive and diverse datasets are critical for developing and refining self-supervised learning methods. We will explore how these methods produce robust neural network models capable of generalizing effectively from sparsely labeled brain activity, significantly enhancing performance in classification tasks and therapeutic interventions. Specifically, we will detail our use of participant-level contrastive learning integrated with transformer architectures, showcasing notable advancements in the efficacy and adaptability of brain-computer interfaces. Learn more: https://www.microsoft.com/en-us/research/video/neural-representation-learning-in-the-wild-toward-generalizable-representations-and-scalable-citizen-science-for-brain-computer-interfaces/