Research and advances in key technologies such as machine learning, computational game theory and autonomy can turn today's AI into AI Fusion — a system that can augment humans and increase quality of life, safety, productivity, and efficiency in meaningful, transformative ways.

AI Fusion focuses on accelerating distributed AI — enabling AI to evolve from today’s highly structured and deterministic, centralized architecture to a more adaptive and pervasive distributed architecture that autonomously fuses AI capability with the enterprise, the edge and AI-infused systems embedded on-platform. AI algorithms would autonomously discover and ‘move to the data,’ processing it at the edge or on-platform in real-time, and fusing the output with AI algorithms in the enterprise or on other platforms at the edge.

The benefits apply to limitless domains, including healthcare, finance, agriculture, transportation, manufacturing, energy, smart cities and the environment. For the Department of Defense and the intelligence community, this innovation will significantly enhance situational awareness and decision-making by fusing information from systems and sensors across multiple domains — from the enterprise to the edge of the battlefield — to maximize mission effectiveness, reduce risk and save lives.

Carnegie Mellon has signature strengths in every domain needed to achieve AI Fusion: AI frameworks and algorithms, AI-infused systems and microelectronics, AI fabric and abstraction and human-AI interaction.

Challenges

• Today’s AI requires extensive data aggregation and engineering to enable AI algorithms. This takes a lot of time.
• Transporting and aggregating stored data from sensors and edge devices to nodes and the cloud requires extensive and persistent high-bandwidth connectivity.

Our Solution: Achieve AI Fusion 

AI Fusion creates an unprecedented advantage in multidomain operations and cross-domain solutions. AI would operate autonomously on-platform and at the edge, enabling relevant data to be processed in real-time with minimal bandwidth and highly dynamic communications.

Integrated Research Thrusts 

Precursors to AI Fusion are already being seen in recent advances in federated learning and microelectronics optimized for neural networks. But truly unlocking the potential of distributed AI for multidomain operations will require integrated research across five critical thrusts and the co-design and development of AI hardware/software to enhance algorithmic agility and enable distributed algorithmic processing and ensembling. The thrust areas are:

AI Frameworks & Algorithms 

Enabling algorithmic agility and distributed processing will require developing new theoretical frameworks and algorithms that extend autonomous discovery and processing of disparate data beyond the current limits of federated learning, information theory, and meta learning. With these advances, the cloud will enable algorithmic mapping and orchestration between the enterprise and varied military platforms and systems operating at the edge. 

AI Fabric & Abstraction Layer 

The co-design and development of an AI fabric is critical to facilitate distributed algorithmic processing and ensembling between the enterprise and the edge. Extensive research into novel mathematical theorems and frameworks, based on stochastic analysis and models of distributed systems, is necessary to ensure the performance, prioritization, scheduling, resource allocation, and security of the new AI algorithms — especially with the dynamic and opportunistic communications associated with military operations in contested environments. 

AI Microelectronics & AI-Infused Systems 

Supporting dynamic autonomous AI processing at the edge and on platform will require extensive research into novel architectures, processing, and connectivity for AI microelectronics. More importantly, extensive research in the co-design and development of AI microelectronics with the AI algorithms and frameworks and the AI fabric is needed to support algorithmic multithreading on a single embedded chip, or AI-infused system or sensor on-platform. It's also necessary to enable scalable training, inferencing, and prediction for military platforms and sensors operating at the edge across multiple operational domains.  

Human-AI Interaction 

Successful AI systems must be intuitive, usable and improve the lives of their human users. Key aspects of human-AI interaction include: model development and model learning; explainabilty: the ability of the user to understand why the AI system has made the decision it has; and visualizing and working with data.    

AI Engineering & Security 

With the exponential increase in AI applications and deployments, extensive research is needed to establish a new 'AI Engineering' discipline for developing resilient, reliable, and secure AI systems. Simply put, AI engineering and security bring confidence in capability — knowing when AI systems are going to work and when to fix them — across the AI Fusion research thrusts. This task is made more difficult as we embrace algorithmic agility and distributed processing, and fuse AI capabilities between the enterprise, the edge, and on-platform AI-infused systems operating across multiple domains.

Convergence Research Across AI and Cyber-Physical Systems

Achieving AI Fusion requires a convergence between the life sciences, physical sciences, computer sciences and engineering to drive transformational research in AI and cyber-physical systems (CPS). A key goal of AI Fusion research is to develop the core system science needed to engineer complex, distributed cyber-physical systems with cognitive capabilities that people can interact with, benefit from, and depend upon across every aspect of their lives. By abstracting from the particulars of a specific application or domain, AI Fusion seeks to reveal cross-cutting fundamental scientific and engineering principles that underpin the integration of AI with cyber- and physical elements across all application sectors. There is also a convergence of AI Fusion technologies and research thrusts focused on smart and connected communities, the internet of things, and advanced wireless networks.