We run Agua AI through an oracle that identifies a vulnerability by using our localization algorithm. It analyzes execution traces of the program running on the specific test cases in the test suite.
We generate a search Space of successfully validate patches found by the defect localiation algorithm and patched by Agua AI.
We use transforms that contains abstract syntax trees. One if these AST’s matches code in the original code. The other AST exemplifies the replacement source for the generated solution. We are able to abstract away application specific details to capture common patch patterns drawn from our open source training sets with this method.