| |
 |
UNIVERSITY OF BUCHAREST
FACULTY OF PHYSICS Guest
2024-11-22 1:55 |
 |
|
|
|
Conference: Bucharest University Faculty of Physics 2007 Meeting
Section: Electricity and Biophysics
Title:
Artificial Neural Networks Modeling
Authors:
Radu Mutihac
Affiliation:
University of Bucharest - Physics Department
E-mail
mutihac@astralnet.ro
Keywords:
Neural networks, artificial intelligence, adaptive systems, cognitive systems.
Abstract:
Models and algorithms have been designed to mimic information processing and knowledge acquisition of the human brain, which are generically called artificial or formal neural networks (ANNs), parallel distributed processing (PDP), neuromorphic, or connectionist models. ANNs are alternative means to symbol programming aiming to implement neurally inspired concepts in artificial intelligence environments (neural computing), whereas cognitive systems attempt to mimic the actual biological nervous systems (computational neuroscience). All conceivable neuromorphic models lie in between and are supposed to be a simplified but meaningful representation of some reality. Mathematical methods approaching various types of ANNs in a unified way and results from linear and nonlinear system control used to obtain learning algorithms are: (i) Tensor products and pseudo-inverses of linear operators; (ii) Convex and nonsmooth analysis; (iii) Control and viability theory; and (iv) Bayesian statistics. Models can be altered externally, by adopting a different axiomatic structure, and internally, by revealing new inside structural or functional relationships. Ranking several neuromorphic models is ultimately carried out based on some measure of performance.
ANNs may be regarded as dynamic systems (discrete or continuous), whose states are the activity patterns, and whose controls are the synaptic weights, which regulate the flux of information between the processing units (adaptive systems controlled by synaptic matrices). Alternatively, ANNs can recognize the state of the environment and act on the environment to adapt to given viability constraints (cognitive systems controlled by conceptual controls). Knowledge is stored in conceptual controls rather than encoded in synaptic matrices, whereas learning rules describe the dynamics of conceptual controls in terms of state evolution in adapting to viability constraints. Neural modeling need no information concerning correlations of input data, rather nonlinear processing units and a sufficiently large number of variable parameters ensure the flexibility to adapt to any relationship between input and output data.
|
|
|
|