Artificial intelligence research and applications in computer vision are countless, especially with the rapid development of convolutional neural networks (CNN). However, like in all fields, designing the architecture of CNNs remains a crucial challenge in the AI domain, requiring expertise, experience, and even intuition from researchers and engineers. Therefore, prediction-based neural architecture search (NAS) focuses on creating a predictor that directly estimates the performance of a candidate CNN architecture, replacing the manual search for CNN designs. Existing predictor-based NAS methods often rely on graph convolutional networks (GCN) as predictors. This study investigates how the adjacency matrix of GCN impairs their performance as NAS methods. To address this issue, we propose the method called redirection of adjacent trails (RATs), which explicitly corrects the problem. Our research ultimately presents RATs-NAS, an efficient and cost-effective predictor-based NAS approach ...

Manually designed convolutional neural networks (CNNs) architectures such as visual geometry group network (VGG), ResNet, DenseNet, and MobileNet have achieved high performance across various tasks, but design them is time-consuming and costly. Neural architecture search (NAS) automates the discovery of effective CNN architectures, reducing the need for experts. However, evaluating candidate architectures requires significant graphics processing unit (GPU) resources, leading to the use of predictor-based NAS, such as graph convolutional networks (GCN), which is the popular option to construct predictors. However, we discover that, even though the ability of GCN mimics the propagation of features of real architectures, the binary nature of the adjacency matrix limits its effectiveness. To address this, we propose redirection of adjacent trails (RATs), which adaptively learns trail weights within the adjacency matrix. Our RATs-GCN outperform other predictors by dynamically adjusting trai...