UOS Lib (Unified Online-learning Systems LIBrary) is an open source library of online learning algorithms for Matlab®. The intention of this library is twofold. On the one hand, it is possible to compare different approaches to online-learning within a common framework, e.g. to see how new approaches rank in comparison to the state-of-the-art. On the other hand, a task at hand can be solved with different learning algorithms to find the most suitable approach. The UOS Lib is focused on the two tasks of classification and regression and consists of several state-of-the-art learning algorithms. Those can be applied to a number of automatically generated learning poblems which are included within in the library and are automatically evaluated. Furthermore, a simple interface to read datasets from files is included, so well known benchmarks can be run as well.
Features:
- Collection of state-of-the-art learning methods: Perceptron, CW, PA, AROW, RLS, GH, IRMA
- Simple interface to implement new learning algorithms
- Approximation structures: Grid-based lookup table, polynomial
- Simple interface to add new approximation structures
- Synthetic reproducable data source
- Interface to dataset-files
- Reproducable noise and data ordering
- Evaluation of: Cumulative loss, training data loss, ground truth loss
- Visualization for 1D- and 2D-datasets
 | Smart Embedded Systems Group |  |
