Method and apparatus for verifying the missing parameter of CNN using test pattern to improve fault tolerance and fluency robustness in extreme conditions for functional safety

摘要

To provide a method for verifying integrity of at least one parameter of a convolutional neural network (CNN) by using at least one test pattern so that an autonomous vehicle can be driven safely from hacker threats.SOLUTION: A method for verifying integrity of parameters of a convolutional neural network (CNN) by using a test pattern added to original input includes, in a computing device, steps of: allowing an adding unit of an integrity verification module 200 to generate extended input by adding a test pattern to original input; allowing the CNN to generate output for verification by applying convolution operations to extended input; and allowing a comparison unit of the integrity verification module to verify the integrity of the parameters of the CNN by determining validity of output for verification by referring to output for reference.SELECTED DRAWING: Figure 2