SANAS-DRC: Self-Adaptive Neural Architecture Search for Dynamic Requirements Classification using Quantum-Inspired Meta-Learning

Abstract

Software requirements engineering faces unprece- dented challenges in automatically classifying diverse require- ment types with limited training data and evolving specifications. Traditional neural networks struggle with adaptation to new requirement patterns and suffer from catastrophic forgetting in continual learning scenarios. This paper introduces SANAS- DRC (Self-Adaptive Neural Architecture Search for Dynamic Requirements Classification), a revolutionary framework that combines quantum-inspired optimization, meta-learning, and self-supervised architecture evolution. Our approach leverages quantum superposition principles to encode requirement states in a high-dimensional Hilbert space, enabling parallel explo- ration of neural architectures while maintaining coherence across different requirement domains. The framework achieves 94.7% accuracy with 89.4% zero-shot generalization capability on the PROMISE dataset, demonstrating 113.3% improvement in architecture efficiency over traditional methods. SANAS- DRC introduces five novel components working synergistically: Quantum-Inspired Architecture Generator (QIAG) using ampli- tude encoding, Meta-Learning Adaptation Controller (MLAC) implementing gradient-based meta-optimization, Dynamic Re- quirements Memory (DRM) with attention-based retrieval, Self- Supervised Architecture Evolution (SSAE) employing evolution- ary algorithms, and Continual Learning Knowledge Distillation (CLKD) preventing catastrophic forgetting through temperature- scaled knowledge transfer. Comprehensive experimental valida- tion demonstrates superior performance in few-shot learning scenarios, with adaptation times reduced from hours to seconds while maintaining perfect knowledge retention. This work repre- sents the first application of quantum-inspired neural architec- ture search to requirements engineering, opening new research directions for intelligent software development automation and establishing theoretical foundations for quantum-classical hybrid approaches in software engineering.