Berlin 2024 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

TT: Fachverband Tiefe Temperaturen

TT 2: Focus Session: Artificial Intelligence in Condensed Matter Physics I (joint session TT/DY)

TT 2.3: Hauptvortrag

Montag, 18. März 2024, 10:30–11:00, H 0104

Disentangling Multiqubit States using Deep Reinforcement Learning — •Marin Bukov — Max Planck Institute for the Physics of Complex Systems

Quantum entanglement plays a central role in modern quantum technologies. It is widely perceived as a proxy for the quantum nature of physical processes and phenomena involving more than one particle. In this talk, we will revisit the problem of disentangling 4-, 5-, and 6-qubit quantum states with the help of machine learning techniques. We use policy gradient algorithms to train a deep reinforcement learning agent which, given access to the pure state of a multiqubit system, has to find the shortest sequence of disentangling two-qubit gates that brings it to a product state. We leverage the agent's interpolation and extrapolation capabilities to learn (approximately) optimal strategies to disentangle Haar-random states that lack any obvious spatial entanglement structure in the computational basis. Analyzing the protocols found by the agent, we show that any 4-qubit state can be prepared using at most 11 CNOT gates. Last, we also demonstrate the robustness of our agent to various sources of stochasticity common for present-day NISQ devices.

Keywords: entanglement; reinforcement learning; deep learning; qubit; quantum control