AI & Data Science

Radiance SMEs are heavily involved in the study of AI focused on advancing our understanding of the field. We create breakthrough AI technologies and capabilities for DARPA and AFRL that build upon the understanding of algorithm design and analysis, mathematical statistics, graph theory, game theory, complexity theory, formal methods, logic and reasoning, evolutionary algorithms, as well as knowledge and ontology engineering. We work closely with major research universities and industry partners in support of our DOD customers.

Radiance delivers capabilities to DOD and intelligence customers such as NASIC and DIA to solve key challenges associated with

• Modern collection platforms to support exponential growth in collected data. Data collection, storage, and dissemination are vital for AI/ML and cloud processing.
• Data visualization dashboards of complex datasets so end-users can interact, explore and rapidly analyze data, supporting data-driven decisions and operational insights
• Data processing capabilities, including support for PED Modernization (MOD) efforts designed to improve all-source data processing.
• Cloud-based scalable, stateful, and streaming time series monitoring tools to rapidly perform data fusion and automated, configurable analysis of time series data at scale.
• Algorithms and production processing pipelines we developed run in Docker containers on customer premises and cloud infrastructure.
• Kubernetes cluster deployment, CI/CD pipelines, review boards and manual testing. Radiance is currently working the Data Management System (DMS) and workflow components for NASIC’S PED MOD.

Radiance is at the forefront of AI research and engineering. We develop algorithms for ML, deep learning (DL), reinforcement learning (RL), natural language processing (NLP), computer vision (CV), and unmanned autonomous systems (UAS). We enable and scale-up AI-driven applications by architecting high-performance cloud-to-edge computing systems leveraging the latest technologies in batch, stream and graph processing of large datasets. We have developed a variety of approaches for automated data labeling and synthetic data generation.

We have developed advanced deep learning algorithms for NLP large language models, image and video classification, object detection, tracking and automated target recognition for a variety of acoustic, optical, and radio-frequency sensors for space, aerial, ground, and oceanographic multi-domain applications. We have experience developing foundational models using Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), Generative Adversarial Networks (GANs) and Autoencoders. We utilize and develop modern NN architectures such as Graph neural networks (GNNs) and Transformer-based architectures.

We develop generative ML models to augment end-to-end M&S frameworks, e.g. scenario generation for wargaming simulations and improvement of RL model sampling and training efficiency. We design GANs for synthetic image and video generation in ATR applications, as well as target detection and tracking for GEOINT and OPIR scenario modeling and analysis SW.

We deliver NLP solutions to the intelligence community using AI and Computational Linguistics techniques. We develop search engines for written MASINT products that go beyond keyword search to incorporate semantic searching, documents similarity assessments, and user context. We are building a semantic suggestion engine to guide analysts through the process of filing complex requirements forms. Our Suggestions tool uses the information users have entered in previous fields to make intelligent suggestions about the current field the user is working on.

We develop explainable and trustworthy ML models for the DOD intelligence community. These models are trained on historical and near-real time data from open sources, SIGINT, HUMINT, GEOINT and MASINT data, integrating a variety of classified and unclassified sources. Our technologies accelerate the analysis of multi-domain ISR data by identifying tracks and anomalous activity through multiple sensor modalities. We help intelligence analysts to quickly generate actionable insights and identify potential threats with confidence and accuracy.

We design and deploy MLOps hosting environments and pipelines maintained by cross-functional teams of SMEs, data scientists, ML engineers and full-stack SW developers. Our teams work in proximity to the end users (analysts, decision-making stakeholders) and follow industry leading project management frameworks. We improve efficiency and reduce complexity by providing robust, scalable, and reliable data-driven pipelines, infrastructure, processes, and tools. We follow industry standards and practices that are designed to ensure the integrity and reliability of ML models and related training, testing and validation data. We create reliable, scalable, and maintainable ML models, while minimizing the risk of security breaches and vulnerabilities during the SW development and data engineering process following integrated strategies of Zero Trust and DevSecOps.