Deep Reinforcement Learning for Robust Dynamical Decoupling

Techniques that suppress the loss of coherence are widely applicable across the fields of quantum information and sensing. Seminal examples of dynamical decoupling protocols effectively protect a single qubit from its environment by applying a sequence of control pulses, often constructed from a small library of pulses. Depending on the complexity of the system, the length of a suitable decoupling sequence can vary greatly, often resulting in a prohibitively large search space. Modern advances in artificial intelligence have demonstrated success in problems of similar or greater depth (e.g., ∼10³⁶⁰ possible move combinations for games of Go). We utilize deep neural networks based on reinforcement learning algorithms to synthesize control sequences for robust dynamical decoupling and suppression of spin-spin interactions, with specific considerations for the nitrogen-vacancy center spin defect in diamond.