|
“Visual” Mathematics Used in a Novel Machine Learning Architecture
|
|
|
Neven Dragojlovic1
*1 Eyeye Sarl, Morocco, Email : eyeye689@gmail.com
|
|
|
|
Abstract
.At some point or another, everyone has surely had the experienced of, staring at mottled surroundings without any thought in mind and finding how one’s visual system started detecting patterns - like faces or animals or objects – in the mottled background. Once those patterns have been detected, moreover, it becomes difficult to observe the mottled surroundings and not perceive the patterns without engaging in active repression. This spontaneous organization of input into recognizable molds starts with just a few patches of visual input that activate a schema that is already present in the memory structure. The visual system then queries the surrounding features to see if they fit into the activated schema, thus constructing a richer and more compelling pattern. Psychology would interpret this process as a Rorschach test that shows what our unconscious mind is primed to perceive at any moment.
This type of experiences suggeststhat, in order for intelligence to emerge, controlled chaos is necessary. Self-feedback and multiple calculation points, each of which follows simple rules, permit the emergence of chaotic attractor-networks, which bring stability to a system, and which, if networked together can create intelligence. This paper describes a computational system that is capable of such a feat, depending only on a mechanical process that does not require thought or consciousness. The systemonly requires local processing units, their associated memory, and simple software that interprets its immediate environment, (that is, the activity of surrounding processing units). In order to be functional, such a system cannot limit itself to the simplest possible case (such as letters or simple geometric shapes), but must be able to process all types of input and form active networks out of it.
The system described in the following article uses a fully parallel pattern-type language that can be used in multiple, easily joined modules, where each module can be used in processing a specific type of information. As it uses simple programs in each computing element, the information is easily integrated and debugged. Complex statistical models, which form the foundation of most current search and recognition algorithms, are not necessary in this system as it automatically uses simple search and recognition strategies at each computing component.
(Based on U.S. Patent 7,426,500 and pending patent US13/117,176)
|
|
|
|
Keywords
:
distributed computing; Cellular parallel architecture; distributed memory; hexagonal framework; networks; chaotic attractors in complex systems; multi-¬‐level up and down information flow; swarm intelligence
|
|
|
|
URL: http://dx.doi.org/10.7321/jscse.v3.n3.69
|
|
|
|
|