Artificial-intelligence reinforcement learning for quantum metrology with adaptive measurements

Quantum metrology aims for measurements of quantum channel (or process) parameters that are more precise than allowed by classical partition noise (shot noise) given a fixed number of input particles. Specifically the imprecision of the process-parameter estimate scales inversely with the square root of the number of particles in the classical domain and up to inverse-linear in the number of particles in the quantum domain by exploiting entanglement between particles. Quantum adaptive-measurement schemes employ entangled-particle inputs and sequential measurements of output particles with feedback control on the channel in order to maximize the knowledge gain from the subsequent particles being sequentially processed. Quantum-adaptive approaches have the advantage that input states are expected to be easier to make experimentally than for non-adaptive schemes. We are interested in devising input states and adaptive feedback control on the processes to beat the standard quantum-measurement limit in real-world scenarios with noise, decoherence and particle losses. As such procedures are difficult to find even in ideal cases, the usual method of clever guessing is inadequate for this purpose. Instead we employ artificial-intelligence machine learning to find adaptive-measurement procedures that beat the standard quantum limit. I will discuss our approaches using reinforcement learning and evolutionary computation to finding procedures for adaptive interferometric phase estimation and show that machine learning has enabled us to find procedures in the ideal case that outperform previously known best cases in the ideal noiseless, decoherence-free, lossless scenario as well as easily devising robust procedures for noisy, decoherent, lossy scenario.