MouseLand/Kilosort

Fast spike sorting with drift correction

What it solves

Kilosort4 is designed to solve the problem of "spike sorting," which is the process of identifying and separating individual neurons' electrical activity (spikes) from large-scale neural recordings. It allows researchers to distinguish which spikes belong to which specific neuron in a high-density electrode probe recording.

How it works

The software uses a GPU-accelerated pipeline to process binary data files and probe configurations. It identifies spikes and clusters them into units based on their waveforms. The process is integrated with the GUI for parameter adjustment and outputs results in a format compatible with Phy, a manual clustering interface used for refining the results.

Who it’s for

It is primarily for neuroscientists and researchers working with high-density neural recordings who need to automate the process of separating individual neuron activity from raw electrical data.

Highlights

  • GPU Acceleration: Requires at least 8GB of GPU RAM (12GB+ recommended) for high-performance processing.
  • Integrated GUI: Provides a user-friendly interface for loading data and running the sorting pipeline.
  • Phy Integration: Automatically sets "good" units based on estimated contamination rates and exports data for manual refinement in Phy.
  • Cross-Platform Support: Fully tested on Windows 10, 11, and Ubuntu 20.04.

Related

  • Project
    online-ml/awesome-online-machine-learning

    The project is a curated list of resources for online machine learning, including courses, blog posts, software, and research papers, organized by topic.

  • Project
    sknetwork-team/scikit-network

    A Python library for machine learning on graphs that uses sparse matrices for memory-efficient network analysis.

  • Project
    JuliaStats/GLM.jl

    A Julia package for fitting linear and generalized linear statistical models, maintained by the JuliaStats group, with stable and development documentation, automated testing, coverage tracking, and a citation DOI.

  • Project
    cursor/mixture-of-kittens

    Mixture‑of‑Kittens (MoK) is a deterministic mixture‑of‑experts megakernel for NVIDIA Blackwell GPUs. It merges compute and inter‑GPU communication into a single CUDA kernel, supports BF16 and MXFP8, and delivers up to 2.37× speed‑ups over existing MoE implementations. Install via pip, configure a `MoKConfig`, create a symmetric‑memory workspace, build a schedule, and call the functional `forward`/`backward` APIs. Designed for large‑scale MoE training on NVL72 hardware.