Influence Of Information Loss On Grey-box System

Information loss is used to represent the consumption of information.In any system with information transmission,the influence of information loss on the system must be considered.We note that the discussion about information loss is mainly in white-box system.At this time,it is not only easy to determine the information loss,but also to avoid the influence of information loss on the system.In this paper,the research object is selected in gray-box system like stream cipher and CNN.In these systems,due to the unknown properties of a considerable part of systems,it is difficult to define the information loss in systems and evaluate the impact it has produced.This paper studies stream cipher and CNN which information loss are hard to determine,and the influence of information loss on them are discussed in detail.We have achieved the following research results:1.A stream cipher based on rule30+cellular automata is designed,and its security is analyzed in detail.The algorithm has good security and scalability,and can support any length of key.2.The information loss in stream cipher is defined,and the relationship between information loss and statistical security is analyzed.By comparing the difference between information loss in stream cipher and block cipher,we come to the key conclusion that a well-designed block cipher can avoid information loss in theory.3.The information loss of convolution neural network is defined,and it is found that the information loss of convolution neural network accumulates rapidly with the development of structure.At the same time,after studying the distribution of information loss in different architectures,we find that the convolution neural network architecture can support a larger depth only when the architecture contains the structure of controlling information loss.