Matrix Relevance Learning: basic concepts and applications in medicine

24 February 2011
Begin time: 
End time: 

Similarity based classification techniques, such as Kohonen's Learning Vector Quantization (LVQ), can be improved significantly by introducing adaptive distance measures. In this talk, the recently developed Matrix Relevance LVQ (MRLVQ) is introduced and discussed in terms of an example application from the medical domain. Here, the aim is the classification of adrenal tumors based on urinary steroid excretion. MRLVQ provides a criterion for the selection of a reduced set of most discriminative bio-markers and facilitates the development of a non-invasive, highly sensitive diagnosis tool. Furthermore, rank-controlled MRLVQ can be used for the discriminative visualization of the data. An outlook on further extensions and applications of MRLVQ will be provided.