Neural network enhanced time-varying parameter estimation via weak measurement

Weak measurement is employed to measure faint signals due to its capability to amplify detection results above technical noise. However, achieving high amplification effects requires accurate adjustment to the experimental system. Estimating unknown time-varying phases, accurately estimating phases, and sensitively perceiving phase changes pose challenges, demanding the system to continuously remain at the appropriate working range. To address this issue, we propose a neural network-based adaptive weak measurement scheme via single-channel light intensity detection. Through machine learning calibrating the experimental system, the reference phase can be dynamically and accurately adjusted, accommodating time-varying phase changes and ensuring the system operates optimally. Compared with traditional dual-channel weak measurement systems, the scheme reduces experimental complexity. Meanwhile, by accurately adjusting the reference phase, the scheme has higher sensitivity and estimation precision compared to the non-modulated scheme. We validate the effectiveness of the scheme in estimating the period and stochastic time-varying phase. The proposed method highlights the advancement of machine learning in weak measurement systems and can also be applied to other quantum-enhanced measurement schemes.