Research On Machine Learning-Assisted Quantum Cryptography System

Different from conventional cryptography which is based on computational complexity,quantum cryptography provides an unconditional secure communication based on the quantum mechanics.As the important parts of quantum cryptography system,quantum key distribution(QKD)and random number generator(RNG)have developed rapidly in theory and practical application.Based on the laws of quantum physics,QKD can in principle provide unconditional security between two legitimate users(Alice and Bob).However,due to the loopholes of imperfect devices,the security of practical QKD systems are vulnerable to various attacks by an evil eavesdropper(Eve).Combined with the decoy-state method,measurement device-independent QKD(MDI-QKD)can resist the loopholes from detector side-channel attacks and multi-photon components in sources.RNGs are extensively applied in the field of cryptography and security communications that require fast and trusted random numbers.To resolve the issue of periodicity of pseudo random number generator,generated with a deterministic algorithm and a provided seed,quantum random number generators(QRNG),which are based on intrinsic indeterministic nature of quantum properties are proposed and demonstrated.The QRNG is the most typical true random number generator.Machine learning(ML),which can utilize a large size of training sample to recognize patterns and discover intricate structures in large data sets.Therefore,ML can be applied to the practical field of quantum cryptography.This dissertation builds a machine learning model onto MDI-QKD system to apply onto the MDI-QKD system for reference frame calibrations.Besides,the dissertation also implements ML into evaluation of QRNGs to judge the quality.The research contents are as follows:1.A long short-term memory network(LSTM)-assisted MDI-QKD system is built to predict out the phase drift between the two users in advance,and actively perform real-time phase compensations,dramatically increasing the key transmission efficiency.Furthermore,corresponding experimental demonstration is carried out over 100 km and 250 km commercial standard single-mode fibers to verify the effectiveness of the approach.2.A optimized LSTM model is applied to dig the correlation between the data generated from a QRNG.Furthermore,its performance is compared with the NIST standard inspection package.Results show that the ML-based scheme can detect inherent correlations among quantum random numbers and exhibits more advantages than traditional NIST standard inspection package.