Skip to main content

GRAPH-MATCHING APPROACHES FOR INVARIANT PERCEPTION

Objective


Theoretical approaches to the functioning of the nervous system are providing a more and more important in the understanding of coactivity between neurons. They are valuable not only in understanding the mechanisms regulating the efficacy of synaptic transmission, but also in a crucial fashion in the neural representation of character entities be they complex, combined or simply relational.

The work achieved in the framework of this project has provided a number of important contributions. It has extended and defined perception theory initially described to a functional reflection on neurobiological mechanisms underlying compositional cognitive functions (language in particular). The contribution of the dynamic connectionist approach to the theory of statistical estimation has also been defined furthur. The work carried out demonstrates and quantifies appreciable advances with particular reference to traditional connectionist approaches and concrete problems of form recognition, thereby providing a source for further applied developments.
The aim of the project is the study of dynamical connectionist models for pattern recognition. Dynamical connectionism is a new theoretical approach to brain function (von der Malsburg 1981, 1985, 1987; von der Malsburg and Bienenstock 1986, 1987; von der Malsburg and Schneider 1986; Binnenstock and von der Malsburg 1987; Binnenstock 1987a, 1987b) whose fundamental tenet is the representation of knowledge in the brain in the form of dynamical connectivity graphs. This representation format is particularly useful for solving problems of invariant pattern recognition. Mathematically, the main underlying computational problem is graph-matching. Biologically, the theory suggests possible roles for co-activity and fast synaptic plasticity in perceptual tasks.

Funding Scheme

CSC - Cost-sharing contracts

Coordinator

ECOLE SUPERIEURE DE PHYSIQUE ET CHIMIE INDUSTRIELLES DE LA VILLE DE PARIS
Address
Rue Vauquelin 10
Paris
France

Participants (4)

Centre National de la Recherche Scientifique (CNRS)
France
Address

91405 Orsay
Max-Planck-Gesellschaft zur Förderung der Wissenschaften eV
Germany
Address
Am Faßberg 11
37077 Göttingen
TECHNICAL UNIVERSITY OF MUNICH
Germany
Address
21,Alte Akademie 16
85350 Freising
United Kingdom Atomic Energy Authority
United Kingdom
Address
353,Harwell
OX11 0RA Didcot - Oxfordshire