A5056660680- Sparse and Compressive Sensing Techniques
- Stochastic Gradient Optimization Techniques
- Microbial Community Ecology and Physiology
- Machine Learning and ELM
- Agriculture, Soil, Plant Science
- Privacy-Preserving Technologies in Data
- Microbial Metabolic Engineering and Bioproduction
- Enzyme Catalysis and Immobilization
- Distributed Sensor Networks and Detection Algorithms
- Microwave Imaging and Scattering Analysis
- Food Quality and Safety Studies
- Lipid metabolism and biosynthesis
- Fermentation and Sensory Analysis
- Domain Adaptation and Few-Shot Learning
- Algal biology and biofuel production
- Wastewater Treatment and Nitrogen Removal
- Biofuel production and bioconversion
- Anaerobic Digestion and Biogas Production
- Forest, Soil, and Plant Ecology in China
- Odor and Emission Control Technologies
- Complexity and Algorithms in Graphs
- Soil Carbon and Nitrogen Dynamics
- Statistical Methods and Inference
- Aluminum toxicity and tolerance in plants and animals
- Constructed Wetlands for Wastewater Treatment
University of North Carolina at Greensboro
2024
University of Connecticut
2020-2022
Huainan Normal University
2017-2020
Universidad Cristiana Autónoma de Nicaragua
2020
Ningbo University
2014-2016
Nanjing Tech University
2010-2011
Cadmium (Cd), a widespread substance with high toxicity and persistence, is known to cause broad range of adverse effects in all living organisms.
Federated learning allows loads of edge computing devices to collaboratively learn a global model without data sharing. The analysis with partial device participation under non-IID and unbalanced reflects more reality. In this work, we propose federated versions adaptive gradient methods - AGMs which employ both the first-order second-order momenta, alleviate generalization performance deterioration caused by dissimilarity population among devices. To further improve test performance,...
Adaptive gradient methods (AGMs) have become popular in optimizing the nonconvex problems deep learning area. We revisit AGMs and identify that adaptive rate (A-LR) used by varies significantly across dimensions of problem over epochs (i.e., anisotropic scale), which may lead to issues convergence generalization. All existing modified actually represent efforts revising A-LR. Theoretically, we provide a new way analyze prove \textsc{Adam} also depends on its hyper-parameter $\epsilon$, has...
The acquisition of iron is important for the pathogenicity bacteria and blood. Three different culture environments (Fe stimulation, blood agar plate normal plate) were used to stimulate Enterobacter cloacae, their respective pathogenicities compared at proteomic, mRNA metabolomic levels.2D-DIGE combined with MALDI-TOF-MS/MS, RT-PCR 1H NMR analyze differential expression levels proteins, metabolites.A total 109 proteins identified by 2D-DIGE mass spectrometry after pairwise comparison within...
Nonconvex sparse learning plays an essential role in many areas, such as signal processing and deep network compression. Iterative hard thresholding (IHT) methods are the state-of-the-art for nonconvex due to their capability of recovering true support scalability with large datasets. Theoretical analysis IHT is currently based on centralized IID data. In realistic large-scale situations, however, data distributed, hardly IID, private local edge computing devices. It thus necessary examine...
In this study, a high salt-tolerant and glyphosate-degrading strain named BZ8 was isolated from activated sludge. According to 16S rDNA sequencing methods, morphological, physiological biochemical analysis, identified as Agrobacterium tumefaciens. The growth capability of A. tumefaciens were investigated the results showed that optimum conditions for glyphosate degradation under 6% NaCl concentration found follows: inoculation size 10% (v/v), incubation temperature 37℃ initial pH 5.0. Salt...
We propose a hard thresholding method based on stochastically controlled stochastic gradients (SCSG-HT) to solve family of sparsity-constrained empirical risk minimization problems. The SCSG-HT uses batch where size is pre-determined by the desirable precision tolerance rather than full reduce variance in gradients. It also employs geometric distribution determine number loops per epoch. prove that, similar latest methods gradient descent or reduction methods, enjoys linear convergence rate....
As data acquisition technologies advance, longitudinal analysis is facing challenges of exploring complex feature patterns from high-dimensional and modeling potential temporally lagged effects features on a response. We propose tensor-based model to analyze multidimensional data. It simultaneously discovers in reveals whether observed at past time points have impact current outcomes. The coefficient, k-mode tensor, decomposed into summation k tensors the same dimension. introduce so-called...
Stochastically controlled stochastic gradient (SCSG) methods have been proved to converge efficiently first-order stationary points which, however, can be saddle in nonconvex optimization. It has observed that a descent (SGD) step introduces anistropic noise around for deep learning and non-convex half space problems, which indicates SGD satisfies the correlated negative curvature (CNC) condition these problems. Therefore, we propose use separate help SCSG method escape from strict points,...