There is a basic question regarding alpha tensor. In game playing it makes sense to train a NN over a large number of self play games where the simulation wins a large number of times, with the intention of using the trained AI for playing a future game. Whereas in algorithm finding, every time there is a win, there is a algorithm and some times it could be new. If a simulation produces a new result it can very well be used as it is w/o further training a AI model. Can someone pl clarify this ?
There is a basic question regarding alpha tensor. In game playing it makes sense to train a NN over a large number of self play games where the simulation wins a large number of times, with the intention of using the trained AI for playing a future game. Whereas in algorithm finding, every time there is a win, there is a algorithm and some times it could be new. If a simulation produces a new result it can very well be used as it is w/o further training a AI model. Can someone pl clarify this ?