Deep Learning spike sorting steps in SpkeInterface
Key Investigators
- Alessio Buccino
- Samuel Garcia
- Liam Paninski
Project Description
Deep learning (DL) has tackled several complex applications and recently several attempts have been made to use DL methods for spike sorting.
Specifically, DL looks promising for signal denoising, spike detection, waveform denoising, and feature extraction.
Some major issues that we would like to solve with this project are to i) reduce/avoid computational time required for training and ii) generalize solutions across different probe designs/configurations
Objectives
Incorporate the following DL methods into SpikeInterface
- DeepInterpolation (from 1)
- spike detection (from 2)
- waveform denoising (from 2)
- autoencoder-based feature extraction
A second major goal is to create a shared repo (e.g. on the GIN platform) for pre-trained network for different steps, probes, and configurations:
- Setup GIN repo with some pre-trained DL networks (e.g. DeepInterpolation for NP1.0)
- Implement in SpikeInterface methods to search and download a pre-trained network
Approach and Plan
Progress and Next Steps
Materials
Background and References
See the following articles for more information:
- “Removing independent noise in systems neuroscience data using DeepInterpolation” (https://www.nature.com/articles/s41592-021-01285-2)
- “YASS: Yet Another Spike Sorter applied to large-scale multi-electrode array recordings in primate retina” (https://www.biorxiv.org/content/10.1101/2020.03.18.997924v1.abstract)