SMS is an open source, extensible and scalable convex solver for a number of machine learning problems cast in the form of regularized risk minimization problem [1]. It is particularly advantageous for optimizing multivariate performance measure [2]. The implementation is "extensible" because the (problem-specific) loss function modules are encapsulated with a common interface for the main optimizer. Thus it is very simple to incorporate solutions to new problems. At present, SMS can solve the following problems:
SMS is implemented with optimized efficiency. It features parallel optimization, which is fuelled by the PETSc and TAO packages.
We also provide data files in PETSc format, which can be parsed and loaded an order of magnitude faster than the conventional LibSVM format.
[1] |
Xinhua Zhang, Ankan Saha, S. V. N. Vishwanathan Smoothing Multivariate Performance Measures Journal of Machine Learning Research (submitted) [PDF] Uncertainty in Artificial Intelligence (UAI), 2011. [PDF]
|
[2] | Thorsten Joachims, A Support Vector Method for Multivariate Performance Measures, ICML, 2005. |
Last modified: 13 August, 2012