Method and apparatus for verifying the integrity of CNN parameters using test patterns to improve fault tolerance and fracture robustness in extreme situations for functional safety

摘要

PROBLEM TO BE SOLVED: To verify the integrity of at least one parameter of a CNN (Concurrental Neural Network) using at least one test pattern, which enables an autonomous vehicle to safely drive from the threat of a hacker. To provide a method. A method for verifying the integrity of a CNN parameter by using a test pattern added to an original input, and in a computing device, an additional unit of an integrity verification module 200 uses the test pattern as an original input. The extended input is generated by the addition, the CNN generates the verification output by applying the convolution operation to the extended input, and the comparison unit of the integrity verification module verifies the verification output by referring to the reference output. The validity of the CNN parameters is verified by determining the validity of the output for use. [Selection diagram] Figure 2