Information Technology Reference
In-Depth Information
Consider again one of the simplest operations of high computational complex-
ity—contour detection of an arbitrary shape, which is executed by the medium in 1-
5 s. Compare this time with the time required for the same operation by a personal
digital computer.
Suppose that the required error of contour detection of the shape is provided by a
grid with 10
3
10
3
points. We assume that, in general, the number of operations
10
6
. The time of floating-point addition by a personal computer
based on the 600 MHz Pentium III processor is ~(2.5-3)
required is ~5
10
9
s. Hence, the
contour detection time is ~10
2
s. That is, on a 10
3
10
3
grid, the contour detection
time by a PC is two orders of magnitude smaller than in the case of a reaction-
diffusion medium. If the contour of a complex shape is detected, which requires
improved accuracy, the situation changes radically. Execution time of this opera-
tion by the medium remains the same, i.e., 1-5 s. In the case of a personal computer,
it increases to 1 s (for a 10
4
10
5
grid).
This example illustrates a fundamental feature of reaction-diffusion media.
Their advantages become the more tangible with more complex tasks.
Note that the detection of the contour of a shape is one of the simplest operations
of high computational complexity. In the case of other elementary operations
performed by media, their advantages become more pronounced.
The speed of reaction-diffusion media is determined by specific chemical
reactions occurring in the medium. A preliminary analysis shows that, apparently,
media can be developed with execution time of elementary operations 10
1
to 1 s,
which will significantly improve their performance.
I would also like to emphasize that the most promising areas of possible
application of the media considered may be technical devices for which the real
time scale is relatively large. Image analysis in medicine and materials science does
not require high execution speed of operations. An autonomous robot moving on
rough terrain can be controlled with reaction time of the control system ~1 s.
10
4
grid) and to 100 s (for a 10
5
8.2.1.4 Multilayered Architecture
The multilayered architecture has not been used so far in the development of the
models of reaction-diffusion information processing devices. Nevertheless, from
general considerations it follows that it offers significant potential for increasing the
performance of these devices.
Multilayered architecture will enable more efficient implementation of biolog-
ically motivated principles of information processing. In a general case multilay-
ered devices should be characterized by:
Processing and compression of information at each level of processing
Transmitting attractors of the previous level, i.e., the results of data compression,
to the next level
This can increase manifold information capabilities of reaction-diffusion
devices.
