Neural network issuing an order?

I am interested in writing certain software that uses machine learning and performs certain actions based on external data.

However, I ran into a problem (it was always interesting for me) - how can I write machine learning software that defines orders or order sequences?

The problem is that, as I understand it, a neural network receives a beam at the inputs and “resembles” a result based on the results of previous trainings. Instantly (well, more or less). Therefore, I am not sure how the “release of orders” can fit into this system, especially when the actions performed by the system affect the system with a certain delay. I'm also a little unsure how to train this thing.

Examples of such a system:
1. Enemy controller first person shooter. As far as I understand, it is possible to implement a neural network controller for a bot that will switch bot behavior strategies (well, assign them priorities) based on some resources (possibly something like health, ammunition, etc.). But I don’t see a way to make a controller of a higher order that could issue a sequence of commands like "go there and then turn left." In addition, bot actions will affect variables that control the behavior of the bot. That is, shooting reduces the amount of ammunition, falling from a height reduces health, etc.
2. Automated market trader. Of course, you can make a system that will try to predict the next market price of something. However, I don’t see how it is possible to create a system that will issue an order in order to buy something, see the trend, and then sell it in order to make a profit / hide losses.
3. The driver of the car. Again, as I understand it, you can create a system that will support the desired motion vector based on data on position / speed / torque and the results of previous training. However, I see no way to force such a system (learn) to perform a sequence of actions.

those. as I understand it, a neural network is technically a matrix - you give it input, it produces an output. But what about generating sequences of actions that can change the environment program?

, ?

P.S. , , , . (.. / ..).

+3
2

, . , - . , , , , "" . , " ", . " ", .

, , . , .

: : -- , ; - , . , ; , . .

+2

, , . - , , (, ) . , , .

+1

Source: https://habr.com/ru/post/1759814/


All Articles