Deep High-Frequency Extrapolation for Neuronal Spike Restoration

Nari Hong, Jaewon Lee, Boil Kim, Han Kyoung Choe, Kyong Hwan Jin, Hongki Kang

Keywords: implicit neural representation, neuronal spike reconstruction, super resolution, high frequency extrapolation, in vitro neuronal culture, in vivo neural activity

TL;DR: Transformer reconstruction of high-frequency and high-resolution spikes from low-passed neuronal signals

Abstract: Recording neuronal activity using multiple electrodes has been widely used for studying functional mechanisms of the brain. However, handling massive amounts of data is still a challenge. In this paper, we propose a novel strategy to restore high-frequency neuronal spikes from small-volume and low-frequency band signals. Inspired by the fact that high-frequency extrapolation is equivalent to super-resolution problems in 2D signals, we applied a Swin transformer to extrapolate high-frequency information from downsampled neuronal signals both in vitro and in vivo. We found that aliasing components of input signals and the spike jittering-based selection of the training batch improved the performance of reconstructing accurate neuronal spikes. As a result, we observed reasonably restored neuronal spiking activity, including the spike timing, waveforms, and network connectivity, even with the $\times 8$ subsampled dataset.

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.

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: Yes

Please Choose The Closest Area That Your Submission Falls Into: Neuroscience and Cognitive Science (e.g., neural coding, brain-computer interfaces)