Recent Trends in Neural Architecture Search Systems

Explosive research has been done on Neural Architecture Search (NAS) to automatically create high- performing neural architectures. The majority of the architectures in use today have been created manually, by human specialists, which is a labor-intensive and fault-prone procedure. This has sparked a rise in interest among researchers in automated neural architecture search methods, which has unavoidably led to the development of a wide variety of automated neural architecture search methods. Choosing the right architecture design has been found to be crucial, and many deep learning advancements result from its direct benefits. On the basis of various search strategies employed in NAS, we offer an insight into the literature on the subject and divide the current NAS works into three primary categories with an emphasis on recent trends in each category. The performance comparison of these categories is performed. Challenges and some future research directions in neural architecture search are outlined.