Prof. David Horn (TAU)

Date: Apr. 26, 2009

Title: Finding structures in data

Abstract: 

Given a set of data-points (instances) in a general parameter-space
(features) one may be interested in finding relationships among the
data (e.g. clustering) or establishing which are the most relevant
parameters (feature selection). These are typical problems in biology
and medicine as well as in many other fields. After an introduction to
preprocessing techniques, we will discuss two novel methods, UFF
(unsupervised feature filtering) and DQC (dynamic quantum clustering),
and exhibit their results on various examples.

UFF ranks features according to their contribution to an
entropy-measure of the eigenvalues of the correlation matrix.
DQC maps data points into states of an Hilbert-space. Based on an
appropriately chosen potential function, it allows the user to trace
the motion of these states in a time-dependent Schrodinger equation
and deduce relationships among data-points accordingly.