Perceived speech decoding and neurophysiological knowledge mining with explainable AI and non-invasive brain activity recordings
Keywords: MEG, Speech decoding, explainable AI, cortical mechanisms of speech processing
TL;DR: XAI powered time-resolved non-invasive neuroimaging to recover cortical mechanisms of speech perception.
Abstract: Explainable artificial intelligence (XAI) is a branch of AI directed at the development of machine learning (ML) solutions that can be comprehended by the human users. Here we use an interpretable and domain-grounded machine learning architecture applied to non-invasive magnetoencephalographic (MEG) data of subjects performing a speech listening task and discover neurophsyologically plausible spatial-temporal neuronal representations of latent sources identified through self-supervised network training process. Achieving high decoding accuracy in the downstream task our solution bridges the gap between high performance and big data-based AI and the classical neuroimaging research and represents a novel knowledge mining platform where the decoding rule can be interpreted using the accepted in electrophysiology terms and concepts which is likely to advance neuroscientific research.
Primary Area: applications to neuroscience & cognitive science
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 9627
Loading