Evgenii Shuranov

ITMO University, Russian Federation, Amsterdam, Netherlands.

Abstract Title: Analysis of the progress of AI models for EEG processing

Biography:

PhD in sound processing, ITMO/HSE Associate professor, Expert in biosignals: EEG and multimodal signal processing for user state estimation.

Research Interest:

Deep neural networks now dominate vision and language, yet electroencephalography (EEG) has lagged because of different montages, data collection methods, environments, individual differences between people, and, lack of consistency in labeling between experiments and datasets. Growing public datasets (such as the TUH EEG corpus) have begun to ease these constraints, enabling large-scale pre-training and transfer learning. • We review recent efforts to build universal EEG embeddings (or Fundamental EEG models) with different deep learning architectures, showing why early deep models often trailed feature-based pipelines rooted in clinical heuristics. • Our latest transformer model, fine-tuned for seizure detection, delivers state-of-the-art performance while reducing adaptation efforts. • We show methods for evaluating the weaknesses of SOTA models and ways to improve them.