Accelint AI Research
Adaptive Training
Adaptive Training is an advanced instructional method that leverages AI to tailor educational experiences to the needs of individual trainees and teams in real-time. This approach dynamically adjusts the content, pace, and complexity of training materials based on continuous assessments of performance with respect to training objectives, preferences, and individual differences. Accelint’s goal is to optimize training efficiency and effectiveness, ensuring that each learner receives the most appropriate and supportive instruction possible.
Key Features of Accelint’s Adaptive Training Research:
- Personalization: Customizes training experiences based on individual trainee profiles, including prior knowledge and the application of required skills.
- Real-Time Feedback: Provides real-time personalized feedback to trainees to help trainees understand mistakes and grasp concepts quickly.
- Data-Driven Adjustments: Utilizes data analytics to monitor individual trainee and team performance and then adapts the training content dynamically.
- Engagement Enhancement: Incorporates interactive and engaging elements tailored to maintain trainee interest and motivation. • Scalability: Enables diverse trainee environments and scales to accommodate large numbers of trainees.
- Generative: Develops AI capabilities to automatically create diverse content (e.g., intelligent agents and scenarios) to enhance training experiences
Learn More about Accelint AI Applications
Simulation
Simulation-Based Training is an instructional approach that uses XR (virtual/augmented /mixed/live) training environments to replicate real-world military scenarios for training purposes. This method includes synthetic environments (fully virtual settings created using computer-generated imagery and software) and live environments (real-world settings enhanced with virtual elements). Accelint’s goal is to provide trainees with an operationally relevant, controlled, and immersive space where they can practice and hone their skills.
Synthetic training environments include representations of terrain, real-entities, and phenomena (ambient light, weather, and other effects). Accelint’s AI solutions are used to automatically generate realistic scenarios, and create & train non-deterministic, flexible software agents that accurately reflect the behaviors of real entities (friendlies, opposing forces, and neutrals).
Accelint is also conducting research to enhance the efficiency and effectiveness of live training. Based on live training data, Accelint is developing AI solutions to identify live events, behaviors, patterns, and performance trends for use in after action reviews (AARs).
Learn More about Accelint AI Applications
Autonomy
Autonomous systems are force multipliers that extend DoD reach and add mass while reducing cost and risk to personnel. Developing and deploying them, though, requires advanced algorithms that can support collaborative operations in high threat, denied environments for extended periods. Our autonomous algorithms are designed to engender operator trust, ensure safe and effective operation, and remain robust to adversarial actions. Accelint has been actively researching, developing, and deploying autonomous unmanned systems for 25 years. As an Artificial Intelligence (AI) company We bring the full range of AI tools to address the varied challenges faced by these systems: deep neural networks for high-speed maneuvers and decision-making such as dogfighting, cognitive systems for higher-level decision-making that can reason tactically like a human, explain their choices, integrate with seamlessly with human formations, and maintain trust; and swarm intelligence for distributed control and coordination of large, multi-domain formations of uncrewed autonomous systems. Accelint is also a leader in the development of DevEthOps, our approach to the development and enforcement of ethical, legal, and societal implications (ELSI) in our autonomy algorithms.
Examples: Accelint’s swarm intelligence architecture, SwarmMATE, was used by NAVAIR and SCO to control a swarm of uncrewed aerial systems employing heterogeneous sensors to search, find, and track multiple moving targets in a high threat, communications and GPS denied environment. The swarm was able to coordinate the required sensor geometries to optimize sensor fusion performance for a variety of passive and active EO/IR/RF/radar sensors to achieve track accuracy while maintain required connectivity in the face of jamming. Accelint’s ISOLATE technology developed under DARPA SquadX and DARPA URSA, combines collaborative ground and air autonomy and human-AI teaming, to enable a single operator to monitor a complex environment with hundreds of individuals by fusing data from multiple fixed and robotic sensors and probing actions to detect and assess potential threat actors and actions. ISOLATE successfully completed 19 separate DARPA field events with live actors under challenging conditions. The final experiment adapted the tools for monitoring a prison environment. ISOLATE pioneered the development of our DevEthOps process for designing ELSI aligned human-AI systems.
Learn More about Accelint AI Applications
Explainable AI
AI can only be a force-multiplier if it is trusted. Accelint is a leader in the research and development of AI technologies, design methods, and evaluation techniques that enable trusted operation.
Accelint is actively researching techniques for high reliability AI reasoning, technologies and displays that explain AI algorithm behavior, and new techniques for the physiological measurement of operators trust.
Example: Accelint’s EpEx is an explanation engine the leverages causal reasoning to explain complex system behavior. EpEx has been applied to medical diagnostics, autonomous system behavior, and adversary behavior recognition in both cyber and Navy domains. EpEx can be easily integrated into new systems and produce explanations in a range of output modes, including human understandable text. Next generation EpEx research will leverage the power of a large language models (LLMs) to bring an expanding breadth of knowledge to explanation problems.
Learn More about Accelint AI Applications
Data Generation
The DoD needs to be able to deliver updated AI models to warfighters faster than ever before. Achieving that requires new techniques for model generation that aren’t limited by collected training data. Accelint is a leader in developing synthetic data generation techniques across a wide range of data types, from synthetic multi-modal and multi-lingual social media that can replicate a city’s worth of dynamic behavior to synthetic imagery that can emulate a high value target in a wide range of poses, backgrounds, and atmospheric conditions.
Example: Accelint’s ImageZero is an AI/ML toolchain for challenging DoD object, event, and activity detection problems with limited available training data, image fidelity, and computing or networking constraints. ImageZero minimizes the need for collection of training images by using 3D models, real world background images, and physics simulations to create a wide range of high fidelity images for training object detectors. ImageZero research is expanding the range of sensor phenomena that can be modeled and developing a robust ‘Synthetic Data Generation as a Service (SDaaS) pipeline that can be integrated into DoD sensor management pipelines.