Cignal, a small business that develops high-performance computing (HPC) and simulation environments to rapidly train Automated Threat Recognition (ATR) systems, announced that it was selected as a presenter at the 2022 Department of Homeland Security (DHS) Science and Technology Directorate’s (S&T) Silicon Valley Innovation Program (SVIP) Demo Week on Machine Learning.
During its presentation, Cignal will discuss and demonstrate its HPC software stack, which facilitates the dynamic training, evaluation, and deployment of next-generation ATR systems for emerging and complex threats. The Cignal HPC Stack permits users to create new screening paradigms, novel objects, and new artificial intelligence (AI) and machine learning (ML) models to protect the public and critical infrastructure against emerging and complex threats. Using this stack, Cignal created the industry’s first volumetric million-bag dataset, which supports advanced X-ray and CT X-ray systems and AI/ML training for large-scale, trillion-parameter ATR models.
The Cignal HPC Stack consists of CignalRay, Checkpoint, Designer, and Cignal Toolkit (CTK) and PyCTK. CignalRay, a patent-pending compute and render engine, generates 2D and 3D-imagery and labels for AI/ML training. Checkpoint is a dynamic training environment that simulates interactions between materials and objects and also allows users to manipulate objects and their properties, creating an unlimited number of dataset variations. To quickly address emerging threats, Designer allows novel firearms, concealments, and explosive designs to be integrated into the training pipeline for rapid detection of threats, which may be too difficult or hazardous to evaluate in real-world environments. CTK and PyCTK provide AI/ML model developers with seamless, interactive command line and Python access to the underlying compute and rendering engine, while enabling convenience in working with massive datasets.
Cignal will be presenting at the 2022 DHS SVIP Demo Week on Machine Learning on Tuesday, May 3, 2022.
AI/ML models rely on large amounts of labeled training data to learn, and generating this data for ATR applications currently is a labor-intensive, manual process. Cignal’s HPC and simulation capabilities eliminate these barriers, allow users to quickly train or test new detection models without the time, cost, or potential hazards associated with conventional approaches.