Title: Noise prediction and reduction of single electron spin by deep-learning-enhanced feedforward control

Abstract: Noise-induced control imperfection is an important problem in applications of diamond-based nano-scale sensing, where measurement-based strategies are generally utilized to correct low-frequency noises in realtime. However, the spin-state readout requires a long time due to the low photon-detection efficiency. This inevitably introduces a delay in noise-reduction process and limits its performance. Here we introduce the deep learning approach to relax this restriction by predicting the trend of noise and compensating the delay. We experimentally implement feedforward quantum control of nitrogen-vacancy center in diamond to protect its spin coherence and improve the sensing performance against noise. The new approach effectively enhances the decoherence time of the electron spin, which enables exploring more physics from its resonant spectroscopy. A theoretical model is provided to explain the improvement. This scheme could be applied in general sensing schemes and extended to other quantum systems.