Are there similar methods of reinforcement learning?

Question

Neural networks

Perhaps the question will sound stupid or strange.

To reward, create a function that will reward or punish the agent. Is it possible to train an agent with a method that is generally similar to gradient descent, but "move" in the direction of the growth of the reward function and change the weights in general in the same way as when descending?

Answer 1

Gradient descent differs from "gradient ascent" only in the sign of the objective function. All these are methods of numerical search for the extremum of functions, which do not differ when searching for the minimum and maximum.