| | |
Summary: Local metrical information:
application to the perceptual magnet effect
Romain Brasselet and Angelo Arleo
CNRS-UPMC Univ. Paris 6, UMR 7102
F75005, Paris, France
romain.brasselet@upmc.fr
ABSTRACT
In this paper, we propose an extension of the recently
introduced metrical information [1]. We give a local
version of it, where the similarity between two events
can depend on the position in the event space. While the
original method demands the equivocation (or
conditional entropy) to be zero and then maximize the
information, we here define an objective function
requiring a trade-off parameter between the
minimization of equivocation and the maximization of
information. We show that the well-known perceptual
magnet [2] effect can be understood as a consequence
of maximizing this objective function in a low-noise
regime. In a simple case with narrow gaussian
|
