Entropy rate test method based on deep learning

Abstract

The invention discloses an entropy rate test method based on deep learning, and belongs to the technical field of information security and random number generators. The problem of high algorithm complexity of an entropy rate test in a high-dimensional system in the test of the random number generator can be solved. According to the technical scheme solving the technical problem, the method comprises the four steps of entropy source data collection and processing, deep learning model training, data testing and entropy rate calculation, the powerful learning ability of a deep learning model is utilized, random data collected by an entropy source is processed, and then the processed random data is used for training the model; and the trained model tests other data to obtain a prediction result, the minimum entropy is calculated according to parameters such as the prediction result, and the minimum entropy is multiplied by a random data rate generated by an entropy source to obtain an entropy rate test result; and the method is applied to entropy rate testing.

Description

technical field [0001] The invention relates to an entropy rate testing method based on deep learning, belonging to the technical fields of information security and random number generators. Background technique [0002] Random number generation is a key technology for large-scale numerical simulation and information security. The non-reproducibility and unpredictability of physical random numbers are very important for information security applications. [0003] An important issue in physical random number generation is the evaluation of the entropy rate, which is the rate at which an entropy source produces uncertainty, or unpredictability. Fast random number generation does not carefully consider the entropy rate of physical entropy sources. Currently, several methods are proposed to evaluate the entropy rate of physical random number generation, such as Lyapunov index using reconstruction attractors, permutation entropy, block entropy of chaotic time series, KS entropy...

AI Technical Summary

Problems solved by technology

When the sample space is too large or the time step is too long, 90B, etc. cannot calculate the result due to the huge time complexity

Other minimum entropy prediction models based on predictors can only detect entropy source samples with specific statistical characteristics, and the complexity of their algorithms explodes with the increase of sample space, which cannot be well applied to other unknown behaviors and entropy evaluation of multivariate sequences

Images