Fuzzy test case adaptive variation method and device based on reinforcement learning

Abstract

The invention discloses a fuzzy test case adaptive variation method and device based on reinforcement learning, and the method comprises the steps of selecting a variation operation group corresponding to a to-be-mutated test case type to carry out the learning of adaptive variation, thereby selecting a variation strategy in a targeted manner, and achieving the variation operation with a finer granularity. According to the present invention, the type information and the variation operation are further adopted to form the context information to be input into a single-step reinforcement learningmodel, the single-step reinforcement learning model converts the selection problems of different variation operations into the selection of different rocker arms in a multi-arm gambling machine problem, and the context-related confidence interval upper bound algorithm LLUCB is used for learning the adaptive variation operation, so that the adaptive learning of the variation operation under different types of scenes is realized, the variation operation capable of obtaining a higher path coverage rate is used for carrying out test case variation, and the efficiency and the quality of the fuzzytest adaptive variation are improved.

Description

technical field [0001] The invention belongs to the field of information security, in particular to a method and device for self-adaptive variation of fuzzy test cases based on reinforcement learning. Background technique [0002] Fuzz testing is an important method in the field of vulnerability mining. At present, most of the popular fuzz testing tools have good vulnerability discovery capabilities, so fuzz testing has received extensive attention in both practice and research. At present, the more popular fuzzers include libFuzzer, AFL and its improved work AFLFast, AFLGo, VUzzer, QSYM, PTFuzz, AGF, TIFF, etc. Although the above tools can effectively find vulnerabilities in many scenarios, they still have the following two limitations: [0003] (1) Existing research work mainly focuses on how to select the most promising parent seeds for mutation, that is, preferentially select seeds that may trigger vulnerabilities or cover new paths, so as to increase the probability of...

AI Technical Summary

Problems solved by technology

[0005] In view of this, the present invention provides an adaptive mutation method for fuzzy test cases based on reinforcement learning, which mainly solves the problem that the current fuzzy test mutation process is relatively blind and the mutation strategy is relatively coarse-grained, and there is no targeted mutation according to the program design of a specific structure type. strategic question

Images