1 results about "Neural network language models" patented technology

The application discloses a context-aware dependency-guided kernel fuzzing test case mutation method and system, and belongs to the technical field of software security and operating system kernel testing. The method comprises the following steps: dynamically collecting and minimizing a high-potential system call sequence set as a training set; modeling the dependency relationship contained in the training set based on a neural network language model, and using the language model to realize context-aware dependency-guided mutation in the mutation stage, so as to help selecting a system call suitable for the current context and establishing an effective state path; alternately switching between exploration and utilization stages, dynamically scheduling mutation operations through an upper bound confidence algorithm to balance the diversity and efficiency of the test, and avoiding falling into a local optimum; and repeating the above steps until the fuzzing test is completed. The application can improve the quality of test cases, trigger the deep code logic of the kernel, and optimize the coverage rate and the efficiency of vulnerability mining.