This paper addresses the efficiency challenge of Neural Architecture Search (NAS) by formulating the task as a ranking problem. Previous methods require numerous training examples to estimate the accurate performance of architectures, although the actual goal is to find the distinction between "good" and "bad" candidates. Here we do not resort to performance predictors. Instead, we propose a performance ranking method (RankNAS) via pairwise ranking. It enables efficient architecture search using much fewer training examples. Moreover, we develop an architecture selection method to prune the search space and concentrate on more promising candidates. Extensive experiments on machine translation and language modeling tasks show that RankNAS can design high-performance architectures while being orders of magnitude faster than state-ofthe-art NAS systems.
Cloud/snow recognition is one application of satellite remote sensing imagery in natural disaster monitoring. Deep learning technology has contributed to the improvement of the performance of cloud/snow recognition. However, deep learning-based methods cannot well balance the performance and efficiency of cloud/snow recognition. In this paper, an augmented multi-dimensional and multi-grained Cascade Forest is proposed for cloud/snow recognition. The multi-dimensional deep forest structure with the representation learning ability allows it to capture the spatial and spectral information of cloud/snow satellite imagery accordingly equipped with good recognition efficiency. Besides, a simple augmentation Random Erasing method is introduced for enhancing the robustness of cloud/snow recognition. The experimental results on the HJ-1A/1B dataset show that the proposed method improves the performance of cloud/snow recognition by extracting spectral information from multi-spectral satellite imagery. In addition, based on the tree-based structure, the proposed method well balances the performance and efficiency of cloud/snow recognition, which can be considered as an alternative to the Neural Network for cloud/snow recognition.INDEX TERMS Cloud/snow recognition, multi-dimensional and multi-grained, Random Erasing, representation learning.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.