Thursday,
September 11, 2008 - 4:00 P.M., NSH 184
Unconventional computing has been an important research topic since it became clear that the power of conventional von Neumann type computers can not grow much further with the same speed as it did in the last decades. One of the many new computational paradigms exploited is cellular neural/nonlinear network (CNN) computing, many times called also as cellular wave computing. This is based on the theory of cellular neural/nonlinear networks (CNN) [1] and is experimentally realized by different physical principles in the architecture of the CNN universal machine (CNN-UM) [2]. This new computational paradigm is also well suited for implementing time-consuming simulations and solving complex problems in statistical physics.
As a first step a fast realistic random number generator (RNG) was developed using the natural noise of the CNN-UM chip. A non-deterministic RNG is obtained by combining the physical properties of the hardware with a chaotic cellular automaton. Using this RNG a broad class of problems of physics can be handled on this computer. These problems are mainly lattice models, which may require random initial conditions or stochastic (Monte Carlo type) dynamics. Two basic problems will be presented: the 2 dimensional site-percolation problem and the two-dimensional Ising model. Both represent a whole class of problems and the algorithms developed could be easily changed for many kindred problems.
In the last part of the presentation an optimization problem will be discussed. A space-variant CNN in which the parameters of all cells can be separately, locally controlled, is the analog correspondent of an Ising type (Edwards-Anderson) spin-glass system. Using the properties of CNN it is shown that one single operation yields a local energetic minimum of the spin-glass system. In such manner a very fast optimization method, similar to simulated annealing, can be built.