mlpack is a recent effort to provide a "swiss army knife" of fast implementations of machine learning methods, with a focus on tree-based algorithms. It has been shown to be faster than other similar toolkits in a recent paper. Some important issues we are focusing on are the breadth of the algorithms offered, the accessibility of mlpack methods from languages other than C++, and the automatic publication of accurate and relevant timing information for mlpack.
Three projects were accepted for our inaugural year in Google Summer of Code and each of these addresses one of the three points listed above.
Addition of two new collaborative filtering methods to improve the breadth of the algorithms mlpack implements (QUIC-SVD and ALS-WR); this will be done by Mudit Raj Gupta.
The development of a system for the automatic generation of bindings for various languages with minimal maintenance requirements; this will be done by Nick Johnston.
The creation of a system for automatic benchmarking of mlpack methods and comparisons with other libraries, to be integrated into our Jenkins build server setup. This project will be done by Marcus Edel.
We are grateful to have been accepted into Google Summer of Code 2013 and look forward to this wonderful opportunity to contribute open-source code into the machine learning world.
By Ryan Curtin, mlpack Organization Administrator
DUNE, the "Distributed and Unified Numerics Environment" is a modular toolbox for solving partial differential equations, like they arise when simulating weather, oil fields, biomechanics, air flow around your car, etc. DUNE was initiated in 2002 and has since evolved to a rather complete toolbox that provides slim templated C++ interfaces allowing efficient use of legacy and/or new libraries.
We are very excited to have two students contributing to DUNE during Google Summer of Code 2013. Miha Čančula will work on a performance testing framework that will allow us to assess the impact on performance of recent changes. Xiaoxue Gong will help us to visualize surfaces, constructed by bisecting triangles, with Paraview.
By Markus Blatt, DUNE Organization Administrator