Evaluating probabilistic and data-driven inference models for fiber-coupled NV-diamond temperature sensors

Area of Science:

• Quantum Sensing
• Spectroscopy
• Machine Learning Applications

Background:

• Optically detected magnetic resonance (ODMR) is a sensitive technique for temperature sensing.
• Accurate temperature inference from ODMR spectra is crucial for various scientific and technological applications.
• Existing methods face challenges in uncertainty quantification and extrapolation capabilities.

Purpose of the Study:

• To evaluate the impact of a probabilistic inference model on temperature uncertainty using ODMR measurements.
• To compare the performance of the probabilistic model against non-parametric and data-driven methods.
• To assess the robustness and generalizability of the probabilistic model for temperature extrapolation.

Main Methods:

• Development and application of a probabilistic feedforward inference model using automatic differentiation.
• Leveraging the temperature dependence of spin Hamiltonian parameters for spectral feature analysis.
• Benchmarking against principal component regression (PCR) and 1D convolutional neural networks (CNN).

Main Results:

• The probabilistic model achieved a prediction uncertainty of ± 1 K across a temperature range of 243 K to 323 K.
• Data-driven methods (PCR, CNN) showed lower uncertainties (up to 0.67 K lower) within the training data range.
• The probabilistic model demonstrated superior performance and robustness when extrapolating beyond the training data range, outperforming PCR and CNN by up to tenfold in uncertainty.

Conclusions:

• Probabilistic inference models offer a robust approach for temperature sensing using ODMR, particularly for extrapolation tasks.
• While data-driven methods excel within their training data range, they lack the generalizability of the probabilistic model.
• The developed model provides reliable temperature measurements with quantified uncertainties, enhancing the applicability of ODMR in diverse environments.