Speaker
Description
Standard intensity-based detectors in electron microscopy measure particles such as photons or electrons individually, thereby collapsing their quantum states and preventing entanglement-enhanced metrology. To overcome this, we are part of a project developing a novel setup that integrates a trapped-ion quantum computer into a transmission electron microscope, thereby transforming the quantum computer into a programmable sensor. In this work, we optimize the control strategy for this setup using classical machine learning. We model the sensing process as a parameterized quantum circuit.
While maximally entangled probes can achieve Heisenberg-limited precision locally, they suffer from periodicity ambiguity when prior knowledge is lacking, thereby binding static strategies to the standard quantum limit. To resolve this, we employ model-aware reinforcement learning and Bayesian updating to adaptively optimize measurement controls. Our neural network agent allocates initial measurements to unambiguously identify the parameter region and subsequently leverages entangled states to refine the estimates with higher precision. We test this setup on different spatially modulated field patterns. In many cases, our adaptive methodology successfully recovers an offset version of the Heisenberg-limited scaling across the full parameter domain without requiring any prior knowledge. This highlights how machine learning can be used to optimize quantum-enhanced sensors.
| Topic | Machine learning techniques for experimental quantum information science |
|---|