V O R T R A G
*********
Oesterreichisches Forschungsinstitut
fuer Artificial Intelligence(OFAI)
Freyung 6/6, A-1010 Wien
Tel.: +43-1-533611-20, Fax: +43-1-5336112-77, Email: sec@ofai.at
-------------------------------------------------------------------
Dr. Lawrence Cayton
Max Planck Institute for Biological Cybernetics, Tuebingen
"Fast similarity search for Bregman divergences"
Nearest neighbor (NN) search is a core ingredient in many algorithms for
machine learning, information retrieval, and elsewhere. Reducing the
complexity of NN search has been extensively studied for decades;
however, most of the research applied only to standard Euclidean
distance or, more generally, metrics. In many modern applications,
including text and image analysis, non-metrics notions of distance have
become popular; for instance, documents are often modeled as probability
distributions, and hence a natural notion of distance is the
KL-divergence (relative entropy).
In this talk, I'll present the first techniques for efficient similarity
search when the notion of distance is given by a Bregman divergence.
Bregman divergences are a broad class of distance-like functions that
have attracted much attention in machine learning in recent years. They
present a challenge for efficient similarity search because they can
behave quite exotically; for example, they can be asymmetric and the
triangle inequality need not hold. These techniques rely heavily on
geometric properties of Bregman divergences which we developed with
convex analysis tools.
*********
Zeit: Donnerstag, 29. Oktober 2009, 18:30 Uhr pktl.
Ort: Oesterreichisches Forschungsinstitut
fuer Artificial Intelligence, OFAI
Freyung 6, Stiege 6, 1010 Wien.
OESTERREICHISCHES FORSCHUNGSINSTITUT
FUER ARTIFICIAL INTELLIGENCE
Univ.-Prof. Ing. Dr. Robert Trappl
*********
Falls Sie keine weiteren Zusendungen wuenschen, schicken Sie uns
bitte eine Antwort-Email mit dem Subject "unsubscribe" an
sec@ofai.at.