Meta-Guided Domain-Aware Neural Architecture Search for Rapid Cross- Domain Adaptation

Neural Architecture Search (NAS) has made its way as a powerful prototype for automating deep neural network design. However, conventional NAS approaches are specific to a single domain and computationally expensive, limiting their scalability across multiple domain environments such as computer vision, healthcare analytics, natural language processing, and edge intelligence. This paper introduces a Meta-Guided Domain- Aware NAS (MD-NAS) framework that forms an integration between meta-learning and domain-conditioned architecture generation. Nevertheless traditional NAS that searches independently for each task, the proposed method learns a shared meta- representation of architectural patterns that dynamically adapts to multi domain characteristics. A domain embedding module captures structural data properties and guides architecture sampling. Experimental results across multi-domain benchmarks demonstrate significant improvements in adaptation speed, generalization performance, and computational efficiency.