In collaboration with the University of Innsbruck (Austria), IQOQI Vienna (Austria) and the Max-Planck-Institute for Quantum Optics in Munich, we realized a proof-of-principle experiment that combines novel concepts from artificial intelligence with the potential of ion trap quantum computers.
In the very first experimental demonstration of a quantum-enhanced reinforcement learning system, we investigate a quantum learning agent in a rapidly changing environment.
In the framework of the projective simulation model of reinforcement learning, we demonstrate a generic speed-up in the agent’s decision-making/deliberation time in a system of two radiofrequency-driven trapped ion qubits.
The quantum algorithm underlying the deliberation process is essentially a Grover-like algorithm, which we implemented efficiently using single-qubit rotations and two-qubit conditional quantum dynamics.
Within experimental uncertainties, our results confirm that the decision-making process of the quantum learning agent is quadratically faster compared to classical learning agents.
This experiment highlights the potential of a scalable ion trap quantum computer in the field of quantum-enhanced learning and artificial intelligence.