Quantum reinforcement learning: the maze problem

Abstract Quantum machine learning (QML) is a young but rapidly growing field where quantum information meets learning. Here, we will introduce new QML model generalising the classical concept of reinforcement to domain, i.e. (QRL). In particular, apply this idea maze problem, an agent has learn optimal set actions in order escape from with highest success probability. To perform strategy optimisation, consider hybrid protocol QRL combined deep neural networks. find that learns both and regimes, also investigate its behaviour noisy environment. It turns out speedup does robustly allow exploit useful at very short time scales, key roles played by coherence external noise. This framework high potential be applied different tasks (e.g. transmission/processing rates error correction) new-generation intermediate-scale (NISQ) devices whose topology engineering starting become crucial control knob for practical applications real-world problems. work dedicated memory Peter Wittek.