Automated Feature Extraction Model Design Based on Neural Architecture Search

The automated feature extraction capability of deep learning models is crucial for improving performance. However, existing feature extraction methods often face challenges such as high computational complexity and significant resource consumption. This study proposes an automated feature extraction model based on neural architecture search (NAS). By designing three core components-feature transformation layer, architecture search layer, and evaluation optimization layer-an efficient feature extraction framework is constructed. A cell-based search space design and an improved reinforcement learning algorithm are employed to achieve automated optimization of model architectures. Experimental results show that the proposed model maintains strong feature extraction capabilities while significantly reducing computational costs and resource consumption, offering a novel approach for optimizing feature extraction in deep learning models.