Ziny Flikop's scientific papers
Home | Car traffic | Conferences | Humanoids | Shuttle foam shedding | Stability of a Queuing System | Stability of a Queuing System (pdf) | Data collection | Data collection (pdf) | Invention | Invention (pdf) | Efficiency of a queuing system. | Efficiency of a queuing System (pdf) | Queuing Network Optimization (pdf) | Queuing Network Optimization | Input Space Decomposition | Cellular Network Design | Cellular Network Management | Cellular Network Management (pdf) | Cellular Networks Design (pdf) | Traffic Routing Optimization (pdf) | Traffic Routing Optimization | Predefined control (pdf) | Input Space Decomposition (pdf) | Self-learning (pdf) | Predefined Control | Self-Learning | About Me | Contact Me

Some thoughts about self-learning systems.

We can divide man-made systems into three broad categories:

  1. They are deterministic systems, behavior of which is strictly predefined by theirs creator. All ancient automatic systems and, up to today fly-by-wire airplanes belongs to this category. These systems are inherently incapable to find solutions to new unforeseen problems.     
  2. They are simple self-learning systems, which are capable of finding via input space decomposition previously unknown solutions for the different problems by using problem-resolving algorithms predefined by a system creator. These systems have limited intelligence and they are inherently incapable to find more advance solutions to the problems then can find creator who uses the same algorithms.
  3. They are humanoid systems which can via self-learning generate problem resolving algorithms more efficient than algorithms that can be provided by a system creator. As a result, these systems have potential to find solutions to the problems that system creator is incapable to find. In other words, intelligence of such systems can be superior to the intelligence of a system creator.  As an example, we can consider a self-learning process demonstrated by a person who is successfully resolving SuDoku puzzles of constantly increasing difficulty. This person starts this process by using simple puzzle resolving algorithms presented in the beginning of every SuDoku book. However, to be capable to find solutions for puzzles of constantly advanced difficulty without external help he/she has persistently try to find by himself/herself more and more advanced puzzle resolving algorithms. As a result, he/she will eventually demonstrate intelligence superior to the intelligence contained in the suggested in the book algorithms. 


In cases when we have more then one self-learning system systems can cooperate as a group by exchanging outcomes of their self-learning and create commonly used database of solutions. It should be noted, that this kind of cooperation could drastically speed up process of evolution of group intelligence.


As we said earlier, systems of a second group cannot exceed intelligence of theirs creator, however, systems of a third group have a potential to do so. 


Enter supporting content here