VORTRAG: Prof.Dr. Miroslav Kubat "MULTI-LABEL HIERARCHICAL CLASSIFICATION"

Wien (OTS) - Perhaps the most common machine-learning task is induction of classifiers from pre-classified training examples described by vectors or attribute values. The classifier's task is then to label a future object with the (single) class it belongs to. In many realistic domains, however, such object belongs to several classes at the same time; moreover there can be thousands of different classes, and they are often inter-related in ways captured by generalization trees or DAG graphs. What we need, then, is a Multi-Label Hierarchical Classifier. This, however, brings about certain difficulties that need to be understood. The talk discusses the problem of performance evaluation in domains of this kind, and then describes a novel system based on the now-popular support-vector machines. Experiments with real-world domains from the field of gene-function prediction then illustrate diverse aspects of this tool.

VORTRAG: Prof.Dr. Miroslav Kubat, Department of Electrical & Computer
Engineering, University of Miami, U.S.A.


"MULTI-LABEL HIERARCHICAL CLASSIFICATION"

Datum: 18.6.2013, um 18:30 Uhr

Ort:
Österreichisches Forschungsinstitut für Artificial
Intelligence (OFAI)
Freyung 6/6/7, 1010 Wien

Url: www.ofai.at

Rückfragen & Kontakt:

Univ.-Prof. Dr. Robert Trappl
Öst. Forschungsinstitut für Artificial Intelligence (OFAI)
Freyung 6/6/7, 1010 Wien
01-5336112-14
robert.trappl@ofai.at

OTS-ORIGINALTEXT PRESSEAUSSENDUNG UNTER AUSSCHLIESSLICHER INHALTLICHER VERANTWORTUNG DES AUSSENDERS | NEF0011