SBIR-STTR Award

RAM Laboratories is proposing Deep Learning for Precise, Automatic and Trusted Code Hardening and Error Removal (DL-PATCHER) which is a combination of a large and diverse source code dataset built from publicly available repositories to be used for training state-of-the-art deep learning algorithms able to produce a model that can automatically generate patches that fix bugs reported by source code analysis tools. In addition, the patch verifier is built utilizing a wide set of heuristics that are able to not only confirm the accuracy of the patch, but also ensures that no major and unexpected feature modifications are made which could significantly impact core functionality. Approved for Public Release | 19-MDA-10270 (18 Nov 19)

When deploying software to systems running in secure environments, it is of upmost importance that everything is done to ensure the software is secure and free of bugs and cyber vulnerabilities. While there exists a large number of tools that scan source code to find potential bugs and vulnerabilities, it is left to developers and subject matter experts (SMEs) to manually fix all of the identified issues. Not only is this a time-consuming process, it is made even worse by the large number of false positives, code that is actually bug free but still flagged by the tool as containing a vulnerability. To address this large technical gap, RAM Laboratories is proposing the Deep Learning for Precise, Automatic and Trusted Code Hardening and Error Removal (DL-PATCHER) solution. Leveraging recent state-of-the-art advances in deep learning, DL-PATCHER is able to use large and diverse code repositories to build neural network models that are able to reason over source code and automatically generate patches that fix bugs and vulnerabilities with astonishingly high accuracy rates. Approved for Public Release | 20-MDA-10643 (3 Dec 20)