/hummus

Spiking neural network simulator

Primary LanguageC++GNU General Public License v3.0GPL-3.0

Logo

  • Hummus is a spiking neural network simulator coded using C++, built first and foremost for neuromorphic computer vision and pattern recognition tasks.

  • Hummus is interfaced with the sepia IO library for event-based devices and it can handle data in the eventstream format for reading and propagating events quickly throughout the network (https://github.com/neuromorphic-paris/sepia for more information on Sepia)

  • The simulator can run both synchronously and asynchronously

Event-based Clock-based
performance easier algorithms
made for neuromorphic platforms neurons updated at every timestep

  • My simulator was made on the fly according to the needs of my phd. Not very optimised but quite convenient to use for pattern recognition.

  • if you have libtorch (C++ frontend of PyTorch) installed on your system, you can even use PyTorch models directly in your spiking neural network. (eg. a logistic regression online classifier is implemented)

  • A matlab toolbox called Hummus Utilities is bundled, in order to easily generate data from popular databases to feed into a network, or to read and perform graphical and statistical analysis on the network output.

  • I've also provided some python scripts to read network output and such

Supported OS

Compilation requires a C++17 compiler and a recent version of CMake. Due to the dependency on the C++17 filesystem library the only supported Operating Systems are:

  • macOS Catalina (10.15+)
  • Ubuntu Disco (19.04+)
  • Debian Buster (10+)

Installation

Some of the following package are purely optional. If these optional packages are found on your system they will be automatically linked to the hummus library.

Install Blaze vectorization library
git clone git@bitbucket.org:blaze-lib/blaze.git
OPTIONAL: install libtorch (easy compatibility with torch models)
cd && wget https://download.pytorch.org/libtorch/nightly/cpu/libtorch-shared-with-deps-latest.zip
unzip libtorch-shared-with-deps-latest.zip

macOS specific

Install Homebrew
/usr/bin/ruby -e "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/master/install)"
Install CMake 3.12+
brew install cmake
OPTIONAL: Install Qt5 (GUI for visualisation)
brew install qt5
OPTIONAL: Install TBB (parallelisation)
cd && brew install tbb

Linux specific

Install CMake 3.12+
sudo apt-get install cmake
Install Qt5 (GUI for visualisation)
sudo apt-get install qt5-default libqt5charts5 libqt5charts5-dev libqt5qml5 qtdeclarative5-dev qml-module-qtcharts qml-module-qtquick2 qml-module-qtquick-controls2
Install TBB (parallelisation)
sudo apt install libtbb-dev

Building Hummus

From the hummus directory run CMake to build Hummus

cmake -S . -B build
cmake --build build

Available build options

  • TORCH: building with libtorch for logistic regression classifier. The location of the TorchConfig.cmake file needs to be specified
cmake -DTORCH=ON -DTorch_DIR=/absolute/path/to/share/cmake/Torch/ -S . -B build
cmake --build build
  • QT: building with Qt for visualisation
cmake -DQT=ON -S . -B build
cmake --build build
  • TBB: building with Intel TBB for parallelisation
cmake -DTBB=ON -S . -B build
cmake --build build

Building for an IDE

an alternative way to build Hummus with an IDE. we'll take the example of XCode on macOS:

cmake -S . -B build -GXcode
cmake --open build

For example, if you're planning on using an IDE such as XCode with libtorch:

cmake -DTORCH=ON -DTorch_DIR=/absolute/path/to/share/cmake/Torch/ -S . -B build -GXcode
cmake --open build