Quantum technology has the potential to enhance our ability to learn about the physical world through both simulation and experiment. With the rapid development of quantum computers aiming towards fault-tolerant operation and long-lived quantum memories, it is natural to try to understand the roles this technology can play, not just in simulation but in processing and learning raw data from our physical world. In this work, we prove that machines with the ability to keep states in quantum memory can learn from exponentially fewer experiments than those with only classical memory in many tasks. Examples include predicting many highly incompatible observables, learning unknown quantum dynamics, and performing quantum principal component analysis. Furthermore, we underscore the power of quantum memory and its accessibility to near-term implementation by experimentally demonstrating an advantage by training a machine learning model that can utilize a noisy quantum memory of at least 40 qubits. These results pave a way towards using quantum sensors, quantum memories, and machine learning to advance physical sciences.
Speaker: Robert Huang
product: Quantum – Research; event: Quantum Summer Symposium 2021; fullname: Robert Huang; re_ty: Publish;