TOM – 12.04.2016

Venue: Biomedicum 1, 5th Floor, Meeting room 7

Timing: 15:30 – 17:00

Speaker: Svetlana Ovchinnikova, PhD Student, Group Anders, FIMM

Metric Learning Algorithms

Many popular machine learning approaches strongly rely on distance/similarity between samples. Yet a pre-defined distance metric is not always relevant for the considered property and an appropriate transformation of the feature space can increase the effectiveness of a machine learning algorithm.

In my presentation, I will give an overview of existing approaches to metric learning and talk about mathematical background behind these algorithms. I will also demonstrate two Matlab packages: LMNN (Large Margin Nearest Neighbours) and MLKR (Metric Learning for Kernel Regression), which can be used for classification and regression problems respectively.

Weinberger, K. Q., Blitzer, J., & Saul, L. K. (2005). Distance metric learning for large margin nearest neighbor classification. In Advances in neural information processing systems (pp. 1473-1480). – http://machinelearning.wustl.edu/mlpapers/paper_files/NIPS2005_265.pdf

Weinberger, K. Q., & Tesauro, G. (2007). Metric learning for kernel regression. In International Conference on Artificial Intelligence and Statistics (pp. 612-619). – http://machinelearning.wustl.edu/mlpapers/paper_files/AISTATS07_WeinbergerT.pdf

Kireeva, N. V., Ovchinnikova, S. I., Kuznetsov, S. L., Kazennov, A. M., & Tsivadze, A. Y. (2014). Impact of distance-based metric learning on classification and visualization model performance and structure–activity landscapes. Journal of computer-aided molecular design, 28(2), 61-73. – http://link.springer.com/article/10.1007/s10822-014-9719-1