Challenges and opportunities for deep learning and Brain signals using EEG

About the speaker: Hubert Banville is a research scientist at InteraXon Inc. in Toronto, Canada. His work focuses on developing machine learning methodology to extract insights from biosignals and generate new real-world applications of neurotechnology. During his PhD at Inria and Université Paris-Saclay, Hubert developed novel deep learning algorithms to leverage the vast amounts of unlabelled and noisy brain activity data generated by out-of-the-lab electroencephalography applications. With a background in biomedical engineering (Polytechnique Montréal), he also previously conducted research on hybrid brain-computer interfaces (INRS, Université du Québec).

Deep learning has revolutionized many research and application domains in the last decade, pushing the boundaries of automated handling of modalities such as images, text and speech. The fields of neuroscience and neuroimaging, though slower in their adoption, also increasingly rely on deep learning for brain data analysis. Over the last few years, deep neural networks have become a valuable tool for processing electroencephalography (EEG) data across domains such as sleep monitoring, clinical diagnosis and brain-computer interfacing.

In this talk, the speaker will give a brief overview of how deep learning has been applied to EEG data, focusing on domain-specific challenges deep learning is particularly well positioned to help with. Specifically, he will present recent examples showing (1) how self-supervised learning can help train deep neural networks on EEG data in the absence of costly expert-provided labels and (2) how attention mechanisms can be designed to improve the handling of real-world EEG in challenging noise conditions.

Get excited!