Hidden Layer LLC is a Washington, DC Metro area company specializing in the development of static software vulnerability detection tools that utilize recent advances in high performance computing and machine learning.
History
Hidden Layer LLC was started in January 2012 and submitted the proposal "A Machine Learning Approach for Source Code Vulnerability Detection" to DARPA's I2O Cyber Fast Track program (CFT).
The proposal was awarded a CFT contract in early February for a period of performance spanning four months. The contract and the static software vulnerability detection system, described in the proposal, were completed in late May (2012). The system's performance demonstrated promising results across three performance measures: accuracy, precision, and recall.
Public Presentations
Our work has been presented in the following venues:
- DARPA I2O's 2014 Cyber Fast Track Demonstration Day
- NYU Poly's 2013 THREADS conference (CSAW 2013)
Current Work
Recent efforts aim to improve the system's robustness and performance. The following techniques have been evaluated and integrated into our static software vulnerability detection system:
Improved training data selection - reduce system training time and the potential for system over-fitting by selecting "the most representative examples" from a large collection of training examples.
Automated Feature Learning - reduces the amount code required to implement the system's feature extraction routines. Provides an added benefit that can reduce the amount of system resources required to train the system.