Building Optimal Neural Architectures using Interpretable Knowledge

AutoBuild proposes a method to construct high-quality neural architectures by assigning importance scores to architecture modules, reducing the need for exhaustive search.