Enhancing spin-based quantum sensors with machine learning

Objectives

1. Investigate and benchmark adaptive protocols for multi-parameter sequential real-time Bayesian experiment design in quantum sensing, through numerical simulations.
2. Investigation of heuristics based on neural networks, developed through model-aware reinforcement learning.
3. Application to the detection on multiple 13C nuclear spins in diamond with a single spin quantum sensor. Test on experimental data.
4. Theoretical/numerical investigation of an optimized protocol to detect nuclear spins outside of the diamond (related to a small molecule, e.g. with 20 nuclear spins).

Description

The detection of magnetic resonance (ESR/NMR) signals from small ensemble of molecules, down to a single molecule, could unlock unprecedented opportunities not only in biomedical diagnosis, but equally importantly in defence, opening the possibility to identify traces of hazardous substances [1].

One of the most promising approaches in this field of nanoscale magnetic resonance (nano-NMR) is based on harnessing the single electron spin associated with a nitrogen-vacancy (NV) centre in diamond, which acts as a nanoscale quantum probe of magnetic signals from the surrounding dipolarly-coupled nuclear spins. Current nano-NMR experiments are limited by long data acquisition times (several days) [2], along with complex analysis performed by experienced physicists, complicating their deployment in applications.

The goal of this PhD project is to exploit model-aware reinforcement learning [3] to optimise pulse sequences to prioritise the measurement settings that yield the most information, reducing the data acquisition time by >2 orders of magnitude. This estimate stems from the observation that the standard scanning protocol for nano-NMR produces a signal with numerous repetitions and regions containing limited information, making it highly inefficient.