In big-data analysis and classification, machine learning can be of much assistance in solving complex problems. Even playing complex games like Go is a task up to machine learning standards. However, the question is, could machine learning be used to find novel protocols and new algorithms for large-scale quantum communication and other related applications?
Machine learning can be used to identify central quantum protocols, including teleportation, entanglement purification, and the quantum repeater. These are important schemes in long-distance quantum communication, and discovering these has been instrumental in shaping the quantum information processing field.
The usefulness of machine learning goes beyond reproducing known protocols, however. The same approach allows a user to find new and improved solutions to problems in long-distance communication, particularly when dealing with asymmetric situations where channel noise and segment distance are not uniform.
The findings of the researchers are based on projective simulations, where a learning agent model is trained with reinforcement learning and decision making in a physically motivated framework. The learning agent is given a universal gate set, and the desired task is specified through a reward scheme.
From a technical perspective, dealing with stochastic environments and reactions is what the learning agent must go through. The researchers used an idea similar to hierarchical skill acquisition. Solutions to subproblems are learned and then reused in the grand scheme of things. This is important when developing long-distance communication schemes, as it can open the way to designing and implementing new quantum networks using machine learning in the future.