It's technically a different algorithm just under the same "monte carlo" name.
However something interesting of note is that since most applications of monte carlo methods rely on PRNGs instead of true RNGs, they are technically deterministic (as given the same PRNG seed you should always get the same result for a given input).
So what this algorithm is doing instead of using a normal PRNG and a separate heuristic, is to instead query the neural network. This works because the neural net is an heuristic over a massive search space which ends up acting like a very poor PRNG that is heavily biased towards specific results based on the training of the neutal net, which in turn ends up looking like a PRNG with a set of heuristics applied.
The important thing to note is that this is a specialisation of MCTS and as such shouldn't technically work for all use cases.
I can't speak for this use case, but in many other places, one isn't even using a PRNG, but rather Quasi-Monte Carlo techniques with no randomness at all. Monte Carlo doesn't need randomness, but a good coverage of the search space.
However something interesting of note is that since most applications of monte carlo methods rely on PRNGs instead of true RNGs, they are technically deterministic (as given the same PRNG seed you should always get the same result for a given input).
So what this algorithm is doing instead of using a normal PRNG and a separate heuristic, is to instead query the neural network. This works because the neural net is an heuristic over a massive search space which ends up acting like a very poor PRNG that is heavily biased towards specific results based on the training of the neutal net, which in turn ends up looking like a PRNG with a set of heuristics applied.
The important thing to note is that this is a specialisation of MCTS and as such shouldn't technically work for all use cases.