Guided Dropout: Improving Deep Networks Without Increased Computation

Deep convolution neural networks are going deeper and deeper.How-ever,the complexity of models is prone to overfitting in training.Dropout,one of the crucial tricks,prevents units from co-adapting too much by randomly drop-ping neurons during training.It effectively improves the performance of deep net-works but ignores the importance of the differences between neurons.To optimize this issue,this paper presents a new dropout method called guided dropout,which selects the neurons to switch off according to the differences between the convo-lution kernel and preserves the informative neurons.It uses an unsupervised clus-tering algorithm to cluster similar neurons in each hidden layer,and dropout uses a certain probability within each cluster.Thereby this would preserve the hidden layer neurons with different roles while maintaining the model’s scarcity and gen-eralization,which effectively improves the role of the hidden layer neurons in learning the features.We evaluated our approach compared with two standard dropout networks on three well-established public object detection datasets.Experimental results on multiple datasets show that the method proposed in this paper has been improved on false positives,precision-recall curve and average precision without increasing the amount of computation.It can be seen that the increased performance of guided dropout is thanks to shallow learning in the net-works.The concept of guided dropout would be beneficial to the other vision tasks.

CuSx-mediated two reaction systems enable biomimetic photocatalysis in CO_(2) reduction with visible light

The performances of heterogeneous catalysts can be effectively improved by optimizing the catalysts via appropriate structure design.Herein,we show that the catalysis of cuprous sulfide can be boosted by constructing the hybrid structure with Cu_(2)S nanoparticles on amorphous CuSx matrix(Cu_(2)S/CuSx).In the photocatalytic CO_(2) reduction under visible light irradiation,the Cu_(2)S/CuSx exhibited a CO production rate at 4.0μmol h-1 that is 12-fold higher than that of the general Cu_(2)S catalyst.Further characterizations reveal that the Cu_(2)S/CuSx has two reaction systems that realize the biomimetic catalysis,involving in the light reaction on the Cu_(2)S nanoparticle-CuSx matrix heterojunctions for proton/electron production,and the dark reaction on the defect-rich CuSx for CO_(2) reduction.The CuSx matrix could efficiently activate CO_(2) and stabilize the split hydrogen species to hinder undesired hydrogen evolution reaction,which benefits the proton-electron transfer to reduce CO_(2),a key step for bridging the two reaction systems.

Harmonic State Space Based Impedance Modeling and Virtual Impedance Based Stability Enhancement Control for LCC-HVDC Systems

Line commutated converter based high-voltage direct-current(LCC-HVDC)transmissions are prone to harmonic oscillation under weak grids.Impedance modeling is an effective method for assessing interaction stability.Firstly,this paper proposes an improved calculation method for the DC voltage and AC currents of commutation stations to address the complex linearization of the commutation process and constructs an overall harmonic state-space(HSS)model of an LCC-HVDC.Based on the HSS model,the closed-loop AC impedances on the LCC-HVDC sending and receiving ends are then derived and verified.The impedance characteristics of the LCC-HVDC are then analyzed to provide a physical explanation for the harmonic oscillation of the system.The effects of the grid strength and control parameters on system stability are also analyzed.To improve the impedance characteristics and operating stability of the LCC-HVDC system,a virtual impedance based stability enhancement control is proposed,and a parameter design method is considered to ensure satisfactory phase margins at both the sending and receiving ends.Finally,simulation results are presented to verify the validity of the impedance model and virtual impedance based stability enhancement control.

Coupling metal-organic frameworks and wood-based carbon for water remediation

Fresh and clean water is highly demanded throughout the world.To effectively address the need,nanomaterials enabled nanotechnology has been explored as a means of more efficient,reliable,and environmentally friendly approach towards water treatment practices.One concern in adopting nanomaterials is how to retrieve them from water body to avoid secondary contamination.In this work,the earth abundant and sustainable wood,e.g.,basswood,was selected and carbonized into porous carbon as host skeleton,and metal-organic frameworks(MOFs),e.g.,MOF-199 with extremely high surface area,were grown throughout all channels in the porous basswood carbon.Targeting the traditional organic pollutant,methyl orange(MO),the combination of MOFs and basswood carbon(MOFs@carbon)demonstrates a remarkable adsorption capacity,which is 243%and 454%higher than basswood carbon and MOF-199,respectively.Such an outstanding adsorption performance originates from that the positively charged carbon pulls MO molecules close to carbon surface,leading to a high MO molecule concentration,and then the concentration gradient drives the MO molecules to be stored inside MOFs,functioning like pockets.These findings highlight the potential application of coupled MOFs and biomass carbon in addressing water remediation.

Systematic study on size and temporal dependence of micro-LED arrays for display applications

Micro-LEDs are one of the most promising candidates for next-generation displays,yet they are inconvenienced by the efficiency reduction induced by the sidewall defects when pursuing further scaled-down device dimensions.We have systematically investigated both the size and temporal dependence of micro-LEDs.Micro-LED arrays with a mesa size ranging from 7 to 100μm were prepared for display purposes.The luminance and external quantum efficiency(EQE)were measured and discussed.Surprisingly,micro-LED arrays with a smaller mesa size exhibit a higher EQE under 100 ns pulse duration operation when compared with longer pulse duration operations.Under certain short-pulsed excitation,a 7×7μm^(2)micro-LED array even exhibits a>20%higher EQE as compared to the direct current(DC)or the long duration pulse operation condition.We thus concluded that the notorious efficiency reduction induced by sidewall defects in small-sized micro-LED arrays could be significantly reduced by applying short-pulse voltages.

DC Current Order Optimization Based Strategy for Recovery Performance Improvement of LCC-HVDC Transmission Systems

For the safe and fast recovery of line commutated converter based high-voltage direct current(LCC-HVDC)transmission systems after faults,a DC current order optimization based strategy is proposed.Considering the constraint of electric and control quantities,the DC current order with the maximum active power transfer is calculated by Thevenin equivalent parameters(TEPs)and quasi-state equations of LCC-HVDC transmission systems.Meanwhile,to mitigate the subsequent commutation failures(SCFs)that may come with the fault recovery process,the maximum DC current order that avoids SCFs is calculated through imaginary commutation process.Finally,the minimum value of the two DC current orders is sent to the control system.Simulation results based on PSCAD/EMTDC show that the proposed strategy mitigates SCFs effectively and exhibits good performance in recovery.

Physical regulation of copper catalyst with a hydrophobic promoter for enhancing CO_(2)hydrogenation to methanol

The hydrogenation of CO_(2)to methanol,which is restricted by water products,requires a selective removal of water from the reaction system.Here,we show that physically combining hydrophobic polydivinylbenzene with a copper catalyst supported by silica can increase methanol production and CO_(2)conversion.Mechanistic investigation reveals that the hydrophobic promoter could hinder the oxidation of copper surface by water,maintaining a small fraction of metallic copper species on the copper surface with abundant Cu^(δ+),resulting in high activity for the hydrogenation.Such a physically mixed catalyst survives the continuous test for 100 h owing to the thermal stability of the polydivinylbenzene promoter.

Lateral shear performance of sheathed post-and-beam wooden structures with small panels

Sheathed post-and-beam wooden structures are distinct from light-wood structures.They allow for using sheathing panels that are smaller(0.91 m×1.82 m)than standard-sized panels(1.22 m×2.44 m or 2.44 m×2.44 m).Evidence indicates that nail spacing and panel thickness determine the lateral capacity of the wood frame shear walls.To verify the lateral shear performance of wood frame shear walls with smaller panels,we subjected 13 shear walls,measuring 0.91 m in width and 2.925 m in height,to a low-cycle cyclic loading test with three kinds of nail spacing and three panel thicknesses.A nonlinear numerical simulation analysis of the wall was conducted using ABAQUS finite element(FE)software,where a custom nonlinear spring element was used to simulate the sheathing-frame connection.The results indicate that the hysteretic performance of the walls was mainly determined by the hysteretic performance of the sheathing-frame connection.When same nail specifications were adopted,the stiffness and bearing capacity of the walls were inversely related to the nail spacing and directly related to the panel thickness.The shear wall remained in the elastic stage when the drift was 1/250 rad and ductility coefficients were all greater than 2.5,which satisfied the deformation requirements of residential structures.Based on the test and FE analysis results,the shear strength of the post-and-beam wooden structures with sheathed walls was determined.

Commutation Failure Mitigation Method Based on Imaginary Commutation Process

The commutation failure(CF) mitigation effectiveness is normally restricted by the delay of extinction angle(EA)measurement or the errors of existing prediction methods for EA or firing angle(FA). For this purpose, this paper proposes a CF mitigation method based on the imaginary commutation process. For each sample point, an imaginary commutation process is constructed to simulate the actual commutation process.Then, the imaginary EA is calculated by comparing the imaginary supply voltage-time area and the imaginary demand voltage-time area, which can update the imaginary EA earlier than the measured EA. In addition, the proposed method considers the impacts of commutation voltage variation, DC current variation, and phase angle shift of commutation voltage on the commutation process, which can ensure a more accurate EA calculation. Moreover, the DC current prediction is proposed to improve the CF mitigation performance under the single-phase AC faults. Finally, the simulation results based on CIGRE model prove that the proposed method has a good performance in CF mitigation.

Noninvasive preimplantation genetic testing in assisted reproductive technology:current state and future perspectives

Invasive genetic screening of pre-implantation embryos via biopsied trophectoderm(TE)cells has been in use for more than 20 years,while its benefits in selecting euploid embryos remain controversial.Recent advances in the ability to process embryonic cell-free DNA(cfDNA)from blastocoel fluid(BF)and spent culture media(SCM)of blastocysts in a manner similar to that of a biopsied TE sample provide a potential alternative holding great promise for obtaining cytogenetic information of the embryos without intrusive biopsy of traditional biopsy-based pre-implantation genetic testing(PGT).Several studies have reported even higher diagnostic accuracy in non-invasive PGT(ni-PGT)than conventional PGT.However,there are still several technical challenges to be overcome before ni-PGT can be accepted as a reliable genomic information source of embryo.In this review,we have summarized the emergence and current state of ni-PGT,and discussed our own perspectives on their limitations and future prospect.There is still a long way to go before truly wide clinical application of ni-PGT.

Quantitative reconstruction of formation paleo-pressure in sedimentary basins and case studies

Paleo-pressure reconstruction in sedimentary basins is one of the most important aspects of hydrocarbon accumulation research.In view of the advantages and disadvantages of the current methods for paleo-pressure research,a new method to reconstruct the paleo-pressure is presented in this paper.According to the geological background,quantitative analyses of the factors that might control overpressure were first conducted to clarify the contributions of each mechanism during different geological periods.Pressure evolution was reconstructed by fluid-compaction modelling with constraints imposed by the paleo-pressures obtained from fluid inclusions or differential stress methods.Determining the mechanisms responsible for overpressures during geological history is the basic prerequisite for paleo-pressure research.Thus,quantitative studies were conducted of the contributions of disequilibrium compaction,gas charging,oil cracking,temperature reduction,and tectonic uplift and subsidence to overpressures.Three case studies of paleo-pressure reconstruction were performed for the Sinian strata in the Sichuan Basin,Ordovician strata in the north uplift in the Tarim Basin and the Permian strata in the Sulige Gas Field in the Ordos Basin,where these three study sites are normally pressured,weakly over-pressured and abnormally low pressured at present,respectively.The new method developed in this paper is very important for the practical reconstruction of the paleopressure in marine strata and ancient strata in deep basins.