Design, build and harness the unique capabilities of a living neural network to engineer hybrid AI architectures with the efficiency, adaptability and resilience of biological systems.

Artificial intelligence (AI) hardware and software architectures have reached unprecedented performance levels through data driven engineering, but this approach has left significant performance gaps. Indeed, in spite of many model-based efforts to replicate the brain’s performance, AI architectures directly inspired by biology (“bio-inspired”) have generally underperformed. At the same time, the highest performing AI algorithms have significant shortcomings, in particular large energy consumption and difficulty learning rapidly from limited data – tasks at which the human brain excels and which is often required in combat situations.

To address this gap, we will establish a neural network dynamics platform that seeks to overcome AI limitations and demonstrates these capabilities under real-world conditions encountered by Armed Services. The platform will enable measurements of information flow in a living neural network and the harnessing of this data to engineer and validate hybrid AI computing architectures – all with the goal of enhancing real-time analysis and decision making in combat scenarios. This will allow us to demonstrate, for the first time, a hybrid artificial neural network (ANN) – Living Neural Network (LNN) architecture for data driven engineering and decision making for defense applications.

We will collaborate with ARL to design and demonstrate a prototype neural network computing architecture mimicking in vivo neural circuits critical for learning and decision making, and we will harness the large multimodal data streams (different types of signals from living cells) from the platform to demonstrate novel hybrid AI architectures for real-time analytics with multivariate data. The collaboration will build upon the UMD capabilities for shaping, characterizing, and manipulating neural network dynamics, as well as ARL capabilities.

It is expected that this first platform for hybrid computing will open research into novel bio-silico architectures that combine the computational advantages of engineered materials with the adaptability and power efficiency of LNNs. One long-term vision is to enable Hybrid Thinking, to build and leverage bio-inspired networks for human autonomy teaming.