Stabilized Neural Dynamics for Behavioral Decoding via Hierarchical Domain Adaptation
Keywords: Unsupervised Domain Adaptation, Brain-Computer Interface, Neural Dynamics, Lyapunov Theory
Abstract: Brain-Computer Interfaces (BCI) have demonstrated significant potential in neural rehabilitation. However, the variability of non-stationary neural signals often leads to instabilities of behavioral decoding, posing critical obstacles to chronic applications. Domain adaptation technique offers a promising solution. Nonetheless, the existing direct adaptation within latent spaces could result in feature deviations. Therefore, developing a stable and efficient alignment framework is crucial for neural decoders.
In this work, we find that dynamical latent features can be extracted from neural dynamics utilizing causal architectures.
We also demonstrate that the process of self-consistent alignment can generate more stable latent features.
Based on these insights, we propose a novel hierarchical domain adaptation (HDA) method for the alignment of dynamical latent features.
Using Lyapunov theory, we further analytically validate the stability of dynamical features, which experimentally exhibit significant enhancements across various datasets.
Our HDA approach effectively addresses the challenge of non-stationary neural signals, thereby potentially improving the reliability of BCIs.
Supplementary Material: zip
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: 3654
Loading