We present in this paper a systematic study on how to morph a well-trained neural network to a new one so that its network function can be completely preserved. We define this as network morphism in this research. After morphing a parent network, the child network is expected to inherit the knowledge from its parent network and also has the potential to continue growing into a more powerful one with much shortened training time. The first requirement for this network morphism is its ability to handle diverse morphing types of networks, including changes of depth, width, kernel size, and even subnet. To meet this requirement, we first introduce the network morphism equations, and then develop novel morphing algorithms for all these morphing types for both classic and convolutional neural networks. The second requirement for this network morphism is its ability to deal with nonlinearity in a network. We propose a family of parametric-activation functions to facilitate the morphing of any continuous non-linear activation neurons. Experimental results on benchmark datasets and typical neural networks demonstrate the effectiveness of the proposed network morphism scheme.
Lake Sayram is an ancient cold water lake locating at a mountain basin in Xinjiang, China. The lake water is brackish, alkaline, unpolluted, and abundant in SO42− and Mg2+. The lacustrine ecosystem of Lake Sayram has been intensely investigated. However, profiles of the microbial communities in the lake remain largely unknown. In this study, taxonomic compositions of the planktonic and sedimentary bacterial communities in Lake Sayram were investigated using 16S rRNA metagenomics. The lacustrine bacterial communities were generally structured by environmental conditions, including the hydrological and physicochemical parameters. Proteobacteria was the dominating phylum. In the lake water, the genera Acinetobacter and Ilumatobacter held an absolute predominance, implying their metabolic significance. In the bottom sediment, biogeochemically significant bacteria and thermophilic or acidothermophilic extremophiles were recovered. In contrast to the planktonic bacteria, an appreciable portion of the sedimentary bacteria could not be classified into any known taxonomic unit, indicating the largely unknown bacteriosphere hiding in the bottom sediment of Lake Sayram.
We propose a generalized regression neural network (GRNN) approach based on grey relational analysis (GRA) and principal component analysis (PCA) (GP-GRNN) to improve the accuracy of density functional theory (DFT) calculation for homolysis bond dissociation energies (BDE) of Y-NO bond. As a demonstration, this combined quantum chemistry calculation with the GP-GRNN approach has been applied to evaluate the homolysis BDE of 92 Y-NO organic molecules. The results show that the ull-descriptor GRNN without GRA and PCA (F-GRNN) and with GRA (G-GRNN) approaches reduce the root-mean-square (RMS) of the calculated homolysis BDE of 92 organic molecules from 5.31 to 0.49 and 0.39 kcal mol−1 for the B3LYP/6-31G (d) calculation. Then the newly developed GP-GRNN approach further reduces the RMS to 0.31 kcal mol−1. Thus, the GP-GRNN correction on top of B3LYP/6-31G (d) can improve the accuracy of calculating the homolysis BDE in quantum chemistry and can predict homolysis BDE which cannot be obtained experimentally.
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.