Learning dynamic quantum circuits for state preparation

Quantum circuits that allow mid-circuit measurements followed by gates conditioned on measurement outcomes are a recently introduced capability of near-term quantum computers. These dynamic quantum circuits (DQCs) can represent some quantum states with long-range entanglement with a depth that does not scale with system size, making them a promising route to realizing interesting states on noisy quantum computers with limited coherence times. However, DQC constructions in prior works rely heavily on knowing the symmetries and structures of the state to be prepared. In this work we show how to variationally learn DQCs to produce target states in cases where such details are not known, thus extending the applicability of DQCs for state preparation. We benchmark our algorithm on paradigmatic states and models and implement the trained DQCs on real quantum hardware.