About the limits of the linear programming
1) What is the problem with linear programming in artificial intelligence?
The problem with linear programming is that the programs, that are made by using this kind of method are unable to pass or leave lines inside the code without attention. This means that regular computer programs are unable to flexible in unexpected situations.
So how the regular program works, when it notices the barrier on the front of the self-driving car. The image-identification program would send the message to the central processing unit that there is a barrier on the route of the vehicle. And then the central processing unit would call the program, what handles the things like barriers.
2) One way to make computing more effective is the increasing number of computers in the structure. So in the independently driving cars would have own computer for every action of that device.
That thing would cause a problem because every movement of the vehicle must be pre-programmed in the vehicle. And the number of sub-programs is very huge. In this case, the vehicle would be under the control of the internal neural network, where each part of this system would have own computer, what controls things like fuel injection, driving direction and another kind of things.
Also, programming actions to those separately operating systems is easier, than making one, large structure. That means that when we are working with the neural networks, the benefit of that system from the program development point of view is that every single computer can be made own programs, and that thing follows the KISSAP (Keep It So Simple As Possible) methodology.
2.1) If the program boards are small, there is less possibility of making errors.
The computer that would control the fuel injection needs only the data, that should it increase or decrease the level of fuel flow, and that data it would get from the system, what controls the driving speed. So if the speed what is required is 80 km/h, the speed control sends the data to the controller of the fuel injection that "keep the fuel injection in this level", and if the speed goes too low, the system orders to inject more fuel to the cylinder. And if the speed is too high, the system orders to decrease the injection. This helps to keep the program bards so small as possible.
Neural network means that fuel injection, gas pedal, lights, and brakes have independently operating controlling computer, what are interacting with each other. When the sensor sees something, the data would send to every computer in that internal network. That structure allows to make many actions at the same time, and it would allow keeping the databases and their structure smaller, which makes them more effective to use.
3) The linear programming weakness is that it cannot apply for orders. That thing has been tried to fix by programming a couple of lines, what those programs must follow if they want to solve the problem.
That thing is a series of IF-ELSE commands like IF(there is something) Make action number one. ELSE make action number two. In this case, the thing, that causes a bigger risk is at first, because it would be a more important thing to solve. But the computer cannot make new connections between databases independently. That ability allows the system to learn independently.
The thing is that this kind of system needs something, which is more flexible than the standard linear code. The system must have some other kind of topology, what it can use to drive the code. Here term "topology" means that when the system starts to drive the code, it starts at the top of the board, and then the computer system would drive the code by line to line, and that is the thing, that would require the new topology.
That topology seems some kind of mind map because it should effectively look for the right actions, and this is the thing, that makes the linear programming difficult to handle in the case of the independent acting robots. If the connection between boards would be in the wrong place, that means that the system would operate the wrong way, because the linear computer program can do only one thing in the one computer or CPU (Central Processor Unit) at the same time.
So how we can make multiple actions at the same time. One thing is simple. We must just increase the number of computers in the structure, and that thing allows those processors to operate many things at the same time. In this case, each system like transmission, brakes, control wheel, and fuel injection would operate own computer. And when the sensor would send the message, it would send that to every computer in the structure at the same time, and that means that the system can react faster and more effectively, as I wrote earlier.
Image: https://www.imcdb.org/v014167.html
Comments
Post a Comment