DISCERN: Rethinking Neural Networks for Energy-Efficient AI

AI built for edge decision-making in cyber, space and beyond

Overview

From research concept to real-world-ready architecture

Aeris-UK worked with a government research programme to develop low-power, scalable Al for autonomous decision-making in constrained environments. Supported by cyber experts Actica Consulting, the project rethinks how neural networks are designed and deployed, helping shape the next generation of intelligent systems built for speed, efficiency and autonomy at the edge.

The Challenge

Making AI work where power and compute can’t stretch.

Modern AI is powerful – but demanding. In environments like cyber defence, space systems or IoT infrastructure, energy and compute are limited – and relying on distant infrastructure introduces delays that edge AI can’t afford. Traditional models are often too heavy, especially where connectivity is patchy or non-existent.

The client needed AI that could operate independently, act fast and consume less – without compromising on intelligence. Spiking Neural Networks (SNNs) showed promise, but training was slow, scalability limited and deployment viability uncertain.

Our Approach

Smarter architecture – not just smaller models.

Instead of shrinking existing models, we went back to first principles. We combined the efficiency of Spiking Neural Networks (SNNs) with new approaches like Liquid Neural Networks (LNNs) and spatially embedded recurrent architectures (seRNNs). These hybrid models were designed for the edge – compact, energy-aware and deployable where traditional AI struggles.

We evaluated them in realistic cyber simulations, measuring energy use, decision accuracy and responsiveness. Crucially, these models can operate directly at the edge, reducing reliance on cloud infrastructure – and with it, the delays caused by remote processing. With input from Actica Consulting, we explored potential use cases, mapping where each approach works best – from cyber defence to space and critical infrastructure.

The Outcome

AI that runs lighter – and thinks faster.

DISCERN has delivered hybrid AI models that significantly reduce energy use while maintaining adaptive performance. These designs show real promise for enabling decision-making in constrained environments, with early results demonstrating clear gains in energy efficiency and overall decision responsiveness.

While current deployment may still require specialist hardware, we are actively exploring how these models can run on standard platforms – a crucial next step toward wider applicability.