Speaker(s): Viktor Gal (Singapore)
The Shogun Machine learning toolbox provides a wide range of unified and efficient Machine Learning (ML) methods. The toolbox seamlessly allows to easily combine multiple data representations, algorithm classes, and general purpose tools. This enables both rapid prototyping of data pipelines and extensibility in terms of new algorithms. We combine modern software architecture in C++ with both efficient low-level computing backends and cutting edge algorithm implementations to solve large-scale Machine Learning problems (yet) on single machines.One of Shogun’s most exciting features is that you can use the toolbox through a unified interface from C++, Python, Octave, R, Java, Lua, C#, Ruby, etc. This not just means that we are independent of trends in computing languages, but it also lets you use Shogun as a vehicle to expose your algorithm to multiple communities. We use SWIG to enable bidirectional communication between C++ and target languages. One of the focus point of the talk is the introduction of the meta-example framework developed as part of the toolbox: a user can write an example use-case for a method using our ‘meta-example’ syntax, that would then be auto-translated into all of the supported languages of the library. This not only allows us to easily scale with the number of new methods in the sense of having the example use-cases of the method being exposed to all of the target languages, but it allows us to have those methods tested throughout of all the languages.Originally focussing on large-scale kernel methods and bioinformatics, the toolbox saw massive extensions to other fields in recent years. It now offers features that span the whole space of Machine Learning methods, including many classical methods in classification, regression, dimensionality reduction, clustering, but also more advanced algorithm classes such as metric, multi-task, structured output, and online learning, as well as feature hashing, ensemble methods, and optimization, just to name a few. Shogun in addition contains a number of exclusive state-of-the art algorithms such as a wealth of efficient SVM implementations, Multiple Kernel Learning, kernel hypothesis testing, Krylov methods, etc. All algorithms are supported by a collection of general purpose methods for evaluation, parameter tuning, preprocessing, serialisation & I/O, etc; the resulting combinatorial possibilities are huge.
(Type: Talk | Track: AI & Machine Learning | Room: Mendel (Ground Floor))
Event Page: http://2017.fossasia.org
Produced by Engineers.SG