Are there engines that can learn to beat stockfish

I remember mark young's games when stockfish almost always drew against stockfish also in FRC and the question is if it is possible to use learning in order to be able to beat deterministic stockfish in every game from a specific FRC srarting position in mark young's games(at least with white).

Note that the learner can use stockfish engine or a different engine and I simply want a proof that something is possible and let start with a proof that beating stockfish at depth 30 plies with white is possible.

For deterministic version I suggest using fixed depth of 30 plies when usually I expect 30 plies to draw even against 40 plies.

Note that learning can be done by the interface(for example decide that the best game so far is the game that you got the biggest evaluation score snd try simply to repeat the best game with one change in one of the moves to see if you can find a new bigger evaluation).

In this case learning against deterministic opponent is going to end with a win or with a failure when you get a game when changing one of the moves does not help.

It may be possible also to learn with a different way.

Yes. Dragon learns through RL and a hefty portion of the training data is from playing against SF14 and lc0. In short, it’s learning to defeat those engines. While it’s getting generally stronger, it’s getting stronger more quickly against those two opponents.