Machine-Learning-Based Design of Quantum Systems for Extreme Sensing

The constant improvement in metrological devices has advanced the frontiers of science. To push the limits of precision, the next generation of experiments will be comprised of highly engi-neered quantum systems. The intrinsic complexity of quantum mechanics often makes engineering quantum systems to perform a specified task a challenging design problem. We explore a new paradigm for the design and control of devices for performing quantum metrology tasks at levels that surpass what is achievable by conventional methods. In this thesis, we present three levels of design philosophy. The first is the periodic modulation of a control parameter that drives the system at a resonant frequency, leading to the amplification of a specific signal. We use the gener-ation of momentum-squeezed particle-pairs from a driven Bose-Einstein condensate as an example. In the second level, we take advantage of a machine learning technique, reinforcement learning, to optimize the control of quantum states and operations. This approach is demonstrated through the design of the matterwave-optic components used for interferometry in a shaken optical lattice. Going one step further to the third level, we apply reinforcement learning in an end-to-end manner to design the whole protocol for a metrology purpose. This idea is illustrated through the design of a device that measures rotational signals, i.e., a gyroscope, in a two-dimensional shaken optical lattice. The design is not constrained to the conventional configuration, and leads to a remarkable gain in sensitivity. These design methodologies open new avenues to discovery with the potential to significantly advance the state-of-the-art quantum sensors.