ML-Based Fuzzing Test Case Generation

The patent replaces the mechanical/manual process of defining formal grammar specifications with a machine-learning-based system. The ML model automatically learns input format patterns from training data and generates test cases without requiring explicit grammar definitions, thereby substituting the complex manual grammar creation process with an automated learning system.

The patent uses existing valid input examples as training data to create a model that can generate new test cases. Instead of creating abstract grammar rules, the system copies patterns from existing valid inputs and applies them to generate new test cases that follow the same structural patterns.

The patent performs preliminary training of the machine-learning model using training test cases and multiple software versions before actual fuzzing. This preliminary action creates a pre-trained model that understands the input format patterns across different versions, enabling faster and more efficient test case generation during the actual fuzzing phase without requiring exhaustive testing of each version individually.

The patent creates a universal machine-learning model that can generate test cases applicable to multiple software versions and different target programs. The model learns general patterns from training data that can be applied across different versions, making the test case generation process universal rather than version-specific, thereby reducing overall testing time and resources.