Project in ML Advanced HT19
Scope of the project (please indicate clearly which tasks account for the desirable grade higher than E)
- Implement the proposed methods (RVM), validate Tipping’s results on a toy problem. We will implement the methods for some of the datasets used in the paper, and some new datasets. As Tipping’s results show a less RVM superiority for classification problems, we limit the report to classification problems.
Datasets to use:
- Banana - 2d
- Breast cancer - 9d
- German - 20d
Link: http://www.raetschlab.org/Members/raetsch/benchmark/
Our own
- Heart
- Ringnorm
- Deploy RVM on selected classification problems, and compare the results and interpretation to SVM. Again, we will use some problems that the paper uses but also include other datasets (for example, problems with higher dimensional input data). Higher grade.
- Investigate scaling properties of RVMs (compare with SVMs) a. Decrease datasets b. See how the difference changes with smaller dataset, (e.g. linearly, exponentially?) c. Discuss complexity