Convolutional algorithms are imitating life. Like living organisms connect genomes the convolutional algorithms connect their program code. Program code is the genome of the computer algorithm. And that makes the convolutional neural networks and algorithms interesting.
They are acting like living organisms. That means the convolutional computer programs are like virtual lifeforms where the organism is the computer code, that operates in the digital memory of the computer.
In mathematics, convolution is the operation that happens between two functions, and that operation forms a new function. In computer sciences and especially in programming convolution is the case. When neural networks, computer programs, or algorithms connect or change their code.
They form new neural networks, computer programs, or algorithms. The transforming networks are the most flexible versions of the convolution networks. The transforming method can use to interconnect two convolution networks. In that method, the convolution network might have transforming subnetworks.
In neural networks convolution means that the neural networks are interacting. And form a new neural network. The thing is that convolution of the algorithms means that the algorithms that are using that method are like living organisms. They can connect their code and create new algorithms. And that thing means that the convolutional neural networks are more flexible and more powerful than ever before.
In the image, you might see the difference between transforming. And traditional convolutional neural networks. In a traditional neural network, the lower-level artificial neuron or data-handling unit is connected only to part of the upper-level or next-stage data handling units. In convolutional neural networks, the data handling units are connected to the entire line of the upper-level data handling units. That thing makes convolutional data handling systems more flexible and powerful than traditional computer systems.
Image: Difference between traditional convolutional and transforming network.
The databases can also form neural networks that can be physical, virtual, or hybrid.
The neural system can be physical, or it can be virtual or hybrid. Learning computer games is an example of a virtual neural system. The hybrid system means that certain parameter activates a series of databases. In the first line of the database, the series determine what the robot should do when it goes as an example into the shop.
In that case, when the robot goes into the shop its sensors activate the database structure or macro database where are orders how to operate in the shop. The first in database structure tells where to go. There is determined what the shelves are looking at and then the robot can walk to shelves. Then the operation in the linear programming structure moves to the next line or next step of the AI.
Then next lines of database structure determine the next operations. When the robot goes into the shop. It starts to use its sensors and select merchandise. If the wanted merchandise is not on the shelf.
The robot starts to search the alternative actions from its databases. The main processing unit sends the query to databases what it should do.
The robot can walk to the clerk and ask for the merchandise. Or it can take the alternative product. There are the parameters of how the system should select the alternative products.
And when the robot is paid for its purchases it goes to the street. There the central processing unit selects the database entirety that involves the street actions and loads it to the RAM-memory of the computer.
The database can be individual or it can be the so-called macro database. The macro database means a database where are multiple sub-databases. The macro-database is the database group that is used for certain situations.
The system can connect or remove connections between the databases and physical systems.
The thing is that the neural network must have the ability to interconnect networks. But it must have also the possibility to remove or cut the connections between neural networks. We can think of this thing by using as an example the sensorial system that uses infrared and normal CCD-cameras for collecting traffic information. In the daytime, the system might use the normal CCD camera for calculating things like cars. When the light conditions turn low.
The system might turn to use infrared cameras for making that operation. Also, the infrared system is suitable to use in fog. So when some parameters have been filled the network can cut off the unnecessary sensors. That removes unusable information from the system. The system can also connect the infrared and regular cameras in some cases.
If that kind of system where normal, infrared, and radar systems are observing airspace there is possible that when one sensor sees something suspicious that activates other sensors that are targeted to that point.
When artificial intelligence operates in laboratories in a fully controlled environment. And even if it operates perfectly, there is a long journey to so-called "street solutions". When robot walks on streets there are many more non-controlled variables than in some laboratories. Somebody can suddenly walk to robot's route. Or something unexpected like somebody getting a heart attack near a robot.
Actions that those robots should do in those situations. Must program in their memory. The thing is that the databases of street operating robots are extremely large. And that thing means that they should connect in the form of a neural network where the system can search right action in each situation fast and trustable. The fact is that the neural network can be physical but it can also be a group of databases.
The databases can be stored in the databank. They might be in series or macro databases. In macro databases, the sub-databases are stored under topics. Topics might be like this: "walking in the street", "going in the shop" and other things like that. When the robot goes into the shop it just selects the database series that is marked as "going in the shop". And then it will start to act like databases are determining, as I wrote earlier.
https://www.quantamagazine.org/will-transformers-take-over-artificial-intelligence-20220310/
https://en.wikipedia.org/wiki/Convolution
Image: https://www.quantamagazine.org/will-transformers-take-over-artificial-intelligence-20220310/
https://interestandinnovation.blogspot.com/
