Min‐Entropy Estimation for Continuous‐Variable Quantum Random Number Generators via Deep Neural Networks

Abstract The entropy source of a quantum random number generator (QRNG) is theoretically unpredictable, but in practice, imperfections in local oscillators and measuring devices introduce classical noise that inevitably contaminates the unpredictability of the quantum entropy source. This makes it a challenge for traditional methods to accurately estimate the min‐entropy of QRNG, which in turn poses potential threats to the practical security of such systems. To address this issue, a novel min‐entropy predictor based on deep neural networks is proposed to solve the min‐entropy estimation problem for continuous‐variable QRNG (CV‐QRNG). First, the process of random number generation is systematically analyzed in CV‐QRNG and the deviation between theoretical and practical randomness is discussed. Next, deep neural networks are used to construct a comprehensive min‐entropy estimation strategy. Finally, extensive entropy estimation tests are conducted on various types of real‐world quantum random number data. Experimental results show that the proposed predictor provides higher accuracy and reliability in min‐entropy estimation for CV‐QRNG, while also improving execution efficiency. In conclusion, the proposed predictor offers a simple and effective method for min‐entropy estimation in CV‐QRNG.