Visual Data Machine Learning Via Generative AI

Synthetaic empowers users to achieve decision dominance with a next-generation, generative artificial intelligence (AI) solution that consistently outperforms traditional AI approaches. Our Rapid Automatic Imagery Categorization (RAIC) technology is a commercial SaaS (TRL-9) AI solution ready for integration into platforms and the full range of DoD I&W and ATR systems. RAIC was developed to revolutionize massive multi-mode search, provide a transformational human-machine interface enabling unprecedented classification and object detection on operational/real-world data, and automate near real-time (NRT) inference across imagery data sets. RAIC was built utilizing modern technology standards that integrates: 1) big data analytics (Johnson et al., 2017), 2) self-supervised AI, 3) computer vision model-building, and 4) proprietary innovations to minimize human intervention in advancement of automated target detection and automated target recognition (ATD/ATR) processes. Synthetaic technology combines artificial neural networks (ANN) and generative adversarial neural network (GAN) techniques pushing the boundaries of self-supervised deep learning science. RAIC has demonstrated military feasibility for ATR with a commercial data pipeline that: 1) ingests all available data with no requirement for manual labeling, 2) activates domain-specific (e.g., geospatial, FMV, etc.) pretrained artificial neural networks (ANNs), and 3) empowers users through the RAIC search engine connected to their data to automate detection and targeting tasks. Proposed Phase I work delivers a prototype with a containerized version of the current commercial TRL-9 cloud capability for an edge- or deployed platform ATR solution. The opportunity is moving RAIC’s transformative ATR processing capabilities upstream closer to the sensor or on an ISR platform while retaining the distributed workforce human-machine interface enabling anyone to rapidly build and deploy AI models and then demonstrate that those models can be pushed deeper to the edge. An important strategic opportunity for SCO, therefore, is demonstrating operational feasibility for a containerized, deployable version of RAIC that enables forward-based processing of massive ISR data (EO or FMV) for any analyst to build and deploy AI models in minutes to search for any target or object and then deep-train an AI model for use on attritable edge devices. Proposed SOW prototype capabilities include: Rapid prototype of RAIC instantiation onto an Azure Edge device (deployable) with ESRI API integration supporting DoD ATR; Demonstrated feasibility of RAIC to provide deployable, on-premises, tactical replication of core user ATR functions being developed for IL-5 and 6 ATO on DoD networks for EO-IR and FMV data; and, Operational demonstration of the Azure Edge RAIC hardware processing ISR data directly downloaded from an ISR platform to the RAIC hardware on the tarmac or staging area of an FMV platform.