Artificial Intelligence and Machine Learning Basic Research

Research AI / ML methodologies to perform object detection on imagery to augment operations. Investigate meeting the challenge of recognition of surrogate targets in S&T test ranges that are not absolute visual representations, using AI capabilities trained on real operational objects. Perform basic research in the area of intelligence support for operations in support of long range precision fires.

The AI Hub is located at Carnegie Mellon University as a consortium of industry, government, and academia focused on building and optimizing the Army's AI and ML initiatives with the goal of accelerating the fielding of capability. The AI Hub will utilize the Army Artificial Intelligence Innovation Institute (A2I2) data and AI/ML algorithms and software tools to investigate AI and ML capabilities to address the Army's unique problems. The AI Hub will focus on research into AI technologies for future application to Army-relevant areas such as, but not limited to, replication of tactical behaviors to enable autonomous capabilities for maneuver, robotics, predictive maintenance, multi-domain Command, Control, Communications, and Computers(C4), network resiliency and cybersecurity, AI-enhanced common operating picture (CoP), intelligent business and process automation, decision support, AI-enabled collaborative data infrastructure platform, medical support and force protection. Will conduct research in distributed AI fabric, algorithms, and human-computer interaction enables operations in multiple Joint Capability Areas (JCA), including command and control, force application, and logistics. The current centralized AI model can be improved with a distributed AI architecture that will: autonomously search for and discover heterogenous data sources; optimize AI processing across dynamic and opportunistic resources; fuse AI capabilities between the enterprise, the edge, and AI-enabled sensors and systems embedded on platform; model the availability and reliability of critical network and computational resources to autonomously adapt and optimize algorithmic processing; and use efficiently distributed learning without the need to move data across the network. No distributed AI solutions currently exist to comprehensively mitigate the identified vulnerabilities. AI2C will conduct foundational research in the ability of distributed AI to address these vulnerabilities to set the conditions for use in Army systems and downstream advanced AI-applications.

Combat Formations require the ability to autonomously maneuver to identify threats and enable friendly forces to disintegrate and exploit enemy forces in the close and deep maneuver areas. This effort researches AI-based, multi-system approaches to aided threat recognition (ATR) using a combination of autonomous air and ground sensors to build a more accurate operating picture when given zone recon missions. ATR and situational awareness is improved through the direct cooperation and autonomous mobility of the sensors.

Foundation models are the bedrock of modern machine learning development. These machine learning models train on vast amounts of data and capture patterns that generalize beyond their training set. This enables the quick development of accurate models across a wide range of tasks and domains through techniques such as few-shot learning and transfer learning. This research seeks to further develop foundation models of various modalities such as language, vision, and segmentation to provide tools and capabilities that extend to solve many problems, including ones that have not yet been identified. These models will include but are not limited to generative methods. Additionally, this research extends to advanced techniques for more effectively adapting existing foundation models (such as those for language, vision, and segmentation) to other domains applicable to the Army. This unlocks more capabilities in both internally developed models as well as the growing set of public and proprietary foundation models developed elsewhere.

Effectively leveraging modern artificial intelligence (AI) and machine learning (ML) techniques for both enterprise and tactical applications requires robust distributed AI capabilities. This research improves these capabilities with a focus on quickly and efficiently training and deploying models across enterprise and tactical systems, federated learning implementations, deploying state-of-the-art AI and ML algorithms onto ruggedized edge hardware and small form-factor devices with computing capabilities, improving robotic autonomous systems and models deployed on robotic platforms, and governing a large portfolio of distributed ML models. As the distributed network of data and AI/ML models grows and becomes more integrated into warfighting functions, it becomes a bigger attack vector for adversaries. In order to keep ongoing AI and ML developments secure, this research also investigates techniques to attack and compromise AI and ML systems as well as to defend them from attacks.

The modern operational environment is complex with vast amounts of available data, but current processes can be improved to more effectively leverage data to generate better decisions and reduce uncertainty. Artificial intelligence (AI) and machine learning (ML) tools have the potential to find useful information in these data, but they need to be able to effectively communicate this to human decision makers, staffs, and operators. This research focuses on the interaction of human and AI systems, especially in high-stakes environments with complex tasks and high uncertainty. As components of this, the research investigates how to make AI more understandable to humans, how to evaluate the outputs of AI and ML, the safety of interactions between humans and robotic or AI systems, how AI and ML impact decision-making, how to effectively integrate AI into current Army processes, how to train users at various technical skill-levels to interact more effectively with AI and ML, how to use AI and ML to process and summarize large amounts of data for human consumption, and how to ethically apply AI to decision making.

Research AI / ML methodologies to perform object detection on imagery to augment operations. Investigate meeting the challenge of recognition of surrogate targets in S&T test ranges that are not absolute visual representations, using AI capabilities trained on real operational objects. Perform basic research in the area of intelligence support for operations in support of long range precision fires.

The AI Hub is located at Carnegie Mellon University as a consortium of industry, government, and academia focused on building and optimizing the Army's AI and ML initiatives with the goal of accelerating the fielding of capability. The AI Hub will utilize the Army Artificial Intelligence Innovation Institute (A2I2) data and AI/ML algorithms and software tools to investigate AI and ML capabilities to address the Army's unique problems. The AI Hub will focus on research into AI technologies for future application to Army-relevant areas such as, but not limited to, replication of tactical behaviors to enable autonomous capabilities for maneuver, robotics, predictive maintenance, multi-domain Command, Control, Communications, and Computers(C4), network resiliency and cybersecurity, AI-enhanced common operating picture (CoP), intelligent business and process automation, decision support, AI-enabled collaborative data infrastructure platform, medical support and force protection. Will conduct research in distributed AI fabric, algorithms, and human-computer interaction enables operations in multiple Joint Capability Areas (JCA), including command and control, force application, and logistics. The current centralized AI model can be improved with a distributed AI architecture that will: autonomously search for and discover heterogenous data sources; optimize AI processing across dynamic and opportunistic resources; fuse AI capabilities between the enterprise, the edge, and AI-enabled sensors and systems embedded on platform; model the availability and reliability of critical network and computational resources to autonomously adapt and optimize algorithmic processing; and use efficiently distributed learning without the need to move data across the network. No distributed AI solutions currently exist to comprehensively mitigate the identified vulnerabilities. AI2C will conduct foundational research in the ability of distributed AI to address these vulnerabilities to set the conditions for use in Army systems and downstream advanced AI-applications.

Combat Formations require the ability to autonomously maneuver to identify threats and enable friendly forces to disintegrate and exploit enemy forces in the close and deep maneuver areas. This effort researches AI-based, multi-system approaches to aided threat recognition (ATR) using a combination of autonomous air and ground sensors to build a more accurate operating picture when given zone recon missions. ATR and situational awareness is improved through the direct cooperation and autonomous mobility of the sensors.

Foundation models are the bedrock of modern machine learning development. These machine learning models train on vast amounts of data and capture patterns that generalize beyond their training set. This enables the quick development of accurate models across a wide range of tasks and domains through techniques such as few-shot learning and transfer learning. This research seeks to further develop foundation models of various modalities such as language, vision, and segmentation to provide tools and capabilities that extend to solve many problems, including ones that have not yet been identified. These models will include but are not limited to generative methods. Additionally, this research extends to advanced techniques for more effectively adapting existing foundation models (such as those for language, vision, and segmentation) to other domains applicable to the Army. This unlocks more capabilities in both internally developed models as well as the growing set of public and proprietary foundation models developed elsewhere.

Effectively leveraging modern artificial intelligence (AI) and machine learning (ML) techniques for both enterprise and tactical applications requires robust distributed AI capabilities. This research improves these capabilities with a focus on quickly and efficiently training and deploying models across enterprise and tactical systems, federated learning implementations, deploying state-of-the-art AI and ML algorithms onto ruggedized edge hardware and small form-factor devices with computing capabilities, improving robotic autonomous systems and models deployed on robotic platforms, and governing a large portfolio of distributed ML models. As the distributed network of data and AI/ML models grows and becomes more integrated into warfighting functions, it becomes a bigger attack vector for adversaries. In order to keep ongoing AI and ML developments secure, this research also investigates techniques to attack and compromise AI and ML systems as well as to defend them from attacks.

The modern operational environment is complex with vast amounts of available data, but current processes can be improved to more effectively leverage data to generate better decisions and reduce uncertainty. Artificial intelligence (AI) and machine learning (ML) tools have the potential to find useful information in these data, but they need to be able to effectively communicate this to human decision makers, staffs, and operators. This research focuses on the interaction of human and AI systems, especially in high-stakes environments with complex tasks and high uncertainty. As components of this, the research investigates how to make AI more understandable to humans, how to evaluate the outputs of AI and ML, the safety of interactions between humans and robotic or AI systems, how AI and ML impact decision-making, how to effectively integrate AI into current Army processes, how to train users at various technical skill-levels to interact more effectively with AI and ML, how to use AI and ML to process and summarize large amounts of data for human consumption, and how to ethically apply AI to decision making.