Consistent Neural Embeddings through Flow Matching on Attractor-like Neural Manifolds
Keywords: Brain-Computer Interface, Neural Decoding, Flow Matching, Dynamical Stability
Abstract: The primary objective of brain-computer interfaces (BCIs) is to establish a direct connection between neural activity and behavioral actions through neural decoders.
Consistent neural representation is crucial for achieving high-performance behavioral decoding over time.
Due to the stochastic variability in neural recordings, existing neural representation techniques yield dynamical instability, leading to the failure of behavioral decoders in few-trial scenarios.
In this work, we propose a novel Flow-Based Dynamical Alignment (FDA) framework that leverages attractor-like ensemble dynamics on stable neural manifolds, which facilitate a new source-free alignment through likelihood maximization.
The consistency of latent embeddings obtained through FDA was theoretically verified based on dynamical stability, allowing for rapid adaptation with few trials.
Further experiments on multiple motor cortex datasets validate the superior performance of FDA.
The FDA method establishes a novel framework for consistent neural latent embeddings with few trials.
Our work offers insights into neural dynamical stability, potentially enhancing the chronic reliability of real-world 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: 2561
Loading