A quantum system, like a person in a noisy room, gradually loses the thread of conversation: its state dissipates into the surroundings. But sometimes, as if suddenly recalling an important thought, it can recover part of the lost information. Physicists call this phenomenon non-Markovian dynamics: the environment's memory of the system's past allows information to flow back.

Scientists set out to amplify this effect by controlling a two-level system (the simplest quantum object, like an atom with two states) using reinforcement learning. This method is akin to training: a computer agent tries different sequences of control pulses and receives a reward for enhancing information return. Comparison with traditional optimal control showed that the standard approach yields only weak and sporadic bursts of return, while the trained agent synchronizes its actions with the environment's 'favorable memory windows' and achieves a significantly stronger effect.

The best result came from the SAC algorithm, while PPO gave slightly less but smoother pulses that are easier to implement experimentally. This means that the memory of a quantum system can be deliberately trained, turning it from a random occurrence into a working tool.

Such an approach could help protect quantum computations from errors and create more resilient quantum devices.