Reproduce “HuiduRep: A Robust Self-Supervised Framework for Learning Neural Representations from Extracellular Recordings” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Feng Cao, Zishuo Feng, Jicong Zhang, Wei Shi

HuiduRep is a self-supervised framework designed for learning robust neural representations directly from extracellular recordings, integrating contrastive learning with a denoising autoencoder and physiologically inspired data augmentations. This approach yielded improved accuracy and precision in spike sorting on diverse datasets, outperforming established methods like KiloSort4 and MountainSort5, particularly under conditions of noise and electrode drift.

Key insights

  • The combination of contrastive learning and an auxiliary denoising task, guided by the DAE, enhances the learned representations' robustness to noise and drift in extracellular recordings.
  • Physiologically inspired data augmentations effectively simulate real-world recording challenges, allowing the model to learn features that generalize well to diverse and challenging conditions.
  • The DAE can serve as a domain alignment tool during inference, reducing the feature space distance between out-of-distribution data and training data, which improves performance on unseen datasets.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.