AdaSAM: Boosting sharpness-aware minimization with adaptive learning rate and momentum for training deep neural networks.

Sharpness aware minimization (SAM) optimizer has been extensively explored as it can generalize better for training deep neural networks via introducing extra perturbation steps to flatten the landscape of deep learning models. Integrating SAM with adaptive learning rate and momentum acceleration, dubbed AdaSAM, has already been explored empirically to train large-scale deep neural networks without theoretical guarantee due to the triple difficulties in analyzing the coupled perturbation step, adaptive learning rate and momentum step. In this paper, we try to analyze the convergence rate of AdaSAM in the stochastic non-convex setting. We theoretically show that AdaSAM admits a O(1/bT) convergence rate, which achieves linear speedup property with respect to mini-batch size b. Specifically, to decouple the stochastic gradient steps with the adaptive learning rate and perturbed gradient, we introduce the delayed second-order momentum term to decompose them to make them independent while taking an expectation during the analysis. Then we bound them by showing the adaptive learning rate has a limited range, which makes our analysis feasible. To the best of our knowledge, we are the first to provide the non-trivial convergence rate of SAM with an adaptive learning rate and momentum acceleration. At last, we conduct several experiments on several NLP tasks and the synthetic task, which show that AdaSAM could achieve superior performance compared with SGD, AMSGrad, and SAM optimizers.

Coptischinensis Franch root rot infection disrupts microecological balance of rhizosphere soil and endophytic microbiomes.

Introduction: The ecological balance of the plant microbiome, as a barrier against pathogens, is very important for host health. Coptis chinensis is one of the important medicinal plants in China. In recent years, Illumina Miseq high-throughput sequencing technology was frequently used to analyze root rot pathogens and the effects of root rot on rhizosphere microorganisms of C. chinensis. But the effects of root rot infection on rhizosphere microecological balance of C. chinensis have received little attention. Methods: In this study, Illumina Miseq high-throughput sequencing technology was applied to analyze the impact on microbial composition and diversity of C. chinensis by root rot. Results: The results showed that root rot infection had significant impact on bacterial α-diversity in rhizome samples, but had no significant effect on that in leaf samples and rhizosphere soil samples, while root rot infection exhibited significant impact on the fungal α-diversity in leaf samples and rhizosphere soil samples, and no significant impact on that in rhizome samples. PCoA analysis showed that the root rot infection had a greater impact on the fungal community structure in the rhizosphere soil, rhizome, and leaf samples of C. chinensis than on the bacterial community structure. Root rot infection destroyed the microecological balance of the original microbiomes in the rhizosphere soil, rhizome, and leaf samples of C. chinensis, which may also be one of the reasons for the serious root rot of C. chinensis. Discussion: In conclusion, our findings suggested that root rot infection with C. chinensis disrupts microecological balance of rhizosphere soil and endophytic microbiomes. The results of this study can provide theoretical basis for the prevention and control of C. chinensis root rot by microecological regulation.

[Identification, biological characteristics, and control of pathogen causing Pinellia ternata soft rot in Hubei province].

This study was designed to identify the pathogen causing soft rot of Pinellia ternata in Qianjiang of Hubei province and screen out the effective bactericides, so as to provide a theoretical basis for the control of soft rot of P. ternata. In this study, the pathogen was identified based on molecular biology and physiological biochemistry, followed by the detection of pathogenicity and pathogenicity spectrum via plant tissue inoculation in vitro and the indoor toxicity determination using the inhibition zone method to screen out bactericide with good antibacterial effects. The control effect of the bactericide against P. ternata soft rot was verified by the leave and tuber inoculation in vitro. The phylogenetic tree was constructed based on the 16 S rDNA, dnaX gene, and recA gene sequences, respectively, and the result showed that the pathogen belonged to the same branch as the type strain Dickeya fangzhongdai JS5. The physiological and biochemical tests showed that the pathogen was identical to D. fangzhongdai, which proved that the pathogen was D. fangzhongdai. The pathogenicity test indicated that the pathogen could obviously infect leaves at 24 h and tubers in 3 d. As revealed by the indoor toxicity test, 0.3% tetramycin, 5% allicin, and 80% ethylicin had good antibacterial activities, with EC_(50) values all less than 50 mg·L~(-1). Tests in tissues in vitro showed that 5% allicin exhibited the best control effect, followed by 0.3% tetramycin and 10% zhongshengmycin oligosaccharide, and their preventive effects were better than curative effects. Therefore, 5% allicin can be used as the preferred agent for the control of P. ternata soft rot, and 0.3% tetramycin and 10% zhongshengmycin oligosaccharide as the alternatives. This study has provided a certain theoretical basis for the control of P. ternata soft rot.

[Effects of biocontrol strain BZJN1 and streptoprofen on physicochemical properties and bacteria structure of rhizosphere soil of Atractylodes macrocephala].

Soil microorganisms are one of the important biological indictors of soil quality and can reflct the comprehensive ecological environment characteristics of the soil. The research of soil microbial diversity is the key to know the ecological functions and balance with soil. In this paper, high-throughput sequencing on PCR-amplified 16 S rRNA gene V3-V4 fragments was used to determine the bacterial diversity in rhizosphere soil of A. macrocephala under the treatment with BZJN1 or streptoprofen. The results showed that there were no significant differences of the bacteria in A. macrocephala rhizosphere soil of the streptoprofen treatment group and the biocontrol BZJN1 treatment group. All the soil bacteria was classified into 25 categories,67 classes, 108 orders, 167 families and 271 generas, except some unidentified bacteria. Proteobacteria(30.7%-34.8%) was the dominant phylum, of which Alphaproteobacteria(16.8%-18.5%) was the dominant subgroup. Compared with the control group, the relative abundance of multiple phylums bacteria in the rhizosphere soil of A. macrocephala was significantly changed in the streptoprofen treatment group and the biocontrol BZJN1 treatment group. In addition, RDA analysis showed that there was connection with different environmental factors and microbial communities. The abundance of the three genera in the rhizosphere soil of A. macrocephala was significantly positively correlated with Invertase, Urease and AP. PICRUSt function prediction results showed that BZNJ1 could enhance some bacterial functions and promote the plant growth. Biocontrol is a new type of green and safety control pest method. BZNJ1 significantly enhances some bacterial functions on the basis of effectively preventing root rot of A. macrocephala and promoting plant growth, and has no significant effect on the soil bacterial community structure. All the results can provide theoretical support for popularization of BZNJ1.

A new approach for analyzing average time complexity of population-based evolutionary algorithms on unimodal problems.

In the past decades, many theoretical results related to the time complexity of evolutionary algorithms (EAs) on different problems are obtained. However, there is not any general and easy-to-apply approach designed particularly for population-based EAs on unimodal problems. In this paper, we first generalize the concept of the takeover time to EAs with mutation, then we utilize the generalized takeover time to obtain the mean first hitting time of EAs and, thus, propose a general approach for analyzing EAs on unimodal problems. As examples, we consider the so-called (N + N) EAs and we show that, on two well-known unimodal problems, leadingones and onemax , the EAs with the bitwise mutation and two commonly used selection schemes both need O(n ln n + n(2)/N) and O(n ln ln n + n ln n/N) generations to find the global optimum, respectively. Except for the new results above, our approach can also be applied directly for obtaining results for some population-based EAs on some other unimodal problems. Moreover, we also discuss when the general approach is valid to provide us tight bounds of the mean first hitting times and when our approach should be combined with problem-specific knowledge to get the tight bounds. It is the first time a general idea for analyzing population-based EAs on unimodal problems is discussed theoretically.