Google Summer of Code 2017 Projects

Google Summer of Code 2017

We are applying to GSoC 2017. Stay tuned for updates. This page is work in progress, based on a copy-paste from last year.

See the follow up blog posts for how it went.

Main focus

This year's GSoC is about improving and showcasing Shogun, rather than extending it (exceptions allowed). We mainly want to recruit new long-term developers.

• Focus on existing algorithms: We want to improve our algorithms - easier use, efficiency, better documentation and more applications - rather than just adding more algorithms.
• Focus on students: As in 2016, we aim to have fewer students - more intense mentoring, interaction between students, blogging and documenting for individual students.
• Focus on application: This year, we would like to use Shogun to solve some real life ML problems in a self-contained project. If you have a cool idea, let us know.

Joint efforts!

In addition to the technical project, all students will:

• add to our example/testing system: http://shogun.ml/examples
• peer-review a fellow student's work
• jointly help our ever growing issue list and work on a release
• contribute to our GSoC blog

Projects

Project Ideas below are roughly ordered by priority and projects in bold type are more likely to happen.

Improving Shogun's infrastructure

These projects do mostly involve software engineering.

• Unified ML interface, plugin-based architecture
• Shogun Detox 2
• SWIG, Matlab & modular interfaces
• Native MS Windows port

Extending Shogun

Note that projects extending Shogun have a lower priority than projects improving Shogun. If algorithms related projects will happen, they are likely to be based around improvements rather than adding new ones.

Algorithms

• Efficient ML: The usual suspects II
• Large-Scale Gaussian Processes
• Approximate kernel methods
• Hip Deep learning
• Fundamental ML: LGSSMs
• Density Estimation in Infinite Dimensional Exponential Families
• Large scale statistical testing
• HMM cleanup and application
• Solver for the KKT System
• Dual coordinate ascent solver for SO-SVM
• LP/QP Framework
• Debiasing & Cluster computing

Framework

• A Kaggle-style pipeline for supervised learning
• MCMC & Stan
• Flexible modelselection 2
• Independent jobs Framework
• Shogun cloud extensions

Other ideas:

We are also open for your ideas: If you have a cool idea for an application or collaboration with another project, let us know! We are explicitly looking for self-contained projects that use Shogun to tackle (or reproduce) a real-life data problem. Here are requirements for such a project.

Low priority projects:

Infrastructure:

• REST interface

Our list of projects is a growing list.