<p>➀ Researchers from TU Graz and partners developed methods to run AI models on IoT devices with ≤4KB memory, enabling localized AI processing for applications like precision positioning; </p><p>➁ Techniques include modular specialized models, subspace configurable networks (SCNs), quantization, and pruning to optimize efficiency while maintaining accuracy; </p><p>➂ Potential applications span industrial automation (drone navigation), automotive security (keyless entry verification), and smart home systems with extended battery life.</p>