[Todos] MARTES 21 DE JUNIO: SEMINARIO IFLP/DEPARTAMENTO DE FÍSICA

[tierno]

ESTIMADOS: 

AGRADECEMOS LA DIFUSIÓN DEL ANUNCIO DEL SEMINARIO DEL CICLO DE COLOQUIOS
Y SEMINARIOS  

QUE SE ORGANIZAN EN EL INSTITUTO DE FÍSICA LA PLATA/ DEPARTAMENTO DE
FÍSICA. 

SE ADJUNTA MATERIAL PARA DIFUNDIR EN CARTELERAS. 

SALUDOS,  

ATTE. 

ADMINISTRACIÓN IFLP 

                                 
 ------------------------------------------------------------------------------------------


SEMINARIOS  DEL IFLP Y DEL DEPARTAMENTO 

MARTES 21 DE JUNIO A LAS 11HS 

AULA CHICA 

TÍTULO:  QUANTUM PATTERN RECOGNITION 

EXPONEN: GIUSEPPE SERGIOLI, HECTOR FREYTES, FEDERICO HOLIK, MARTIN BOSYK
/UNIVERSITY OF CAGLIARI - UNIVERSITY OF LA PLATA

 

RESUMEN: 

 We introduce a new framework for describing pattern recognition tasks
by means of the mathematical language of density matrices. 

In recent years, many efforts to apply the quantum formalism to
non-microscopic contexts have been made and, in this direction,
important contributions in the areas of pattern recognition and image
understanding have been provided. Even if these results seem to suggest
some possible computational advantages of an approach of this sort, an
extensive and universally recognized treatment of the topic is still
missing. 

The natural motivations which have led to use quantum states for the
purpose of representing patterns are 

i) the possibility to exploit quantum algorithms to boost the
computational intensive parts of the classification process, 

ii) the possibility of using quantum-inspired algorithms for solving
classical problems more effectively. 

In our work, firstly we provide a one-to-one correspondence between
patterns, expressed as n-dimensional feature vectors (according to the
standard pattern recognition approach), and pure density operators (i.e.
points in the n-dimensional Bloch hypersphere) called "density
patterns". By using this correspondence, we give a representation of the
well-known Nearest Mean classifier (NMC) in terms of quantum objects by
defining an "ad hoc" Normalized Trace Distance (which coincides with the
Euclidean distance between patterns in the real space). 

Consequently, we have found a quantum version of the discriminant
function by means of Pauli components, represented as a plane which
intersects the Bloch sphere. 

This first result suggests future potential developments, which consist
in finding a quantum algorithm able to implement the normalized trace
distance between density patterns with a consequent significative
reduction of the computational complexity of the process. 

But the main result we show consists in introducing a purely quantum
classifier (QC), which has not any kind of classical counterpart,
through a new definition of "quantum centroid". The convenience of using
this quantum centroid lies in the fact that it seems to be more
informative than the classical one because it takes into account also
information about the distribution of the patterns. 

As a consequence, the main implementative result consists in showing how
this quantum classifier performs a significative reduction of the error
and an improvement of the accuracy and precision of the algorithm with
respect to the NMC (and also to other commonly used classifiers) on a
classical computer. 

The behaviors of QC and NMC on different datasets will be shown and
compared. 

 
------------ próxima parte ------------
Se ha borrado un adjunto en formato HTML...
URL: <http://mail.fisica.unlp.edu.ar/pipermail/todos/attachments/20160614/f30f53f6/attachment-0001.html>
------------ próxima parte ------------
A non-text attachment was scrubbed...
Name: 21.JUNIO.doc
Type: application/msword
Size: 130048 bytes
Desc: no disponible
URL: <http://mail.fisica.unlp.edu.ar/pipermail/todos/attachments/20160614/f30f53f6/attachment-0001.doc>