Optimization in Machine Learning:a Distribution-Space Approach

We present the viewpoint that optimization problems encountered in machine learning can often be interpreted as minimizing a convex functional over a function space,but with a non-convex constraint set introduced by model parameterization.This observation allows us to repose such problems via a suitable relaxation as convex optimization problems in the space of distributions over the training parameters.We derive some simple relationships between the distribution-space problem and the original problem,e.g.,a distribution-space solution is at least as good as a solution in the original space.Moreover,we develop a numerical algorithm based on mixture distributions to perform approximate optimization directly in the distribution space.Consistency of this approximation is established and the numerical efficacy of the proposed algorithm is illustrated in simple examples.In both theory and practice,this formulation provides an alternative approach to large-scale optimization in machine learning.

Boosting Adversarial Training with Learnable Distribution

In recent years,various adversarial defense methods have been proposed to improve the robustness of deep neural networks.Adversarial training is one of the most potent methods to defend against adversarial attacks.However,the difference in the feature space between natural and adversarial examples hinders the accuracy and robustness of the model in adversarial training.This paper proposes a learnable distribution adversarial training method,aiming to construct the same distribution for training data utilizing the Gaussian mixture model.The distribution centroid is built to classify samples and constrain the distribution of the sample features.The natural and adversarial examples are pushed to the same distribution centroid to improve the accuracy and robustness of the model.The proposed method generates adversarial examples to close the distribution gap between the natural and adversarial examples through an attack algorithm explicitly designed for adversarial training.This algorithm gradually increases the accuracy and robustness of the model by scaling perturbation.Finally,the proposed method outputs the predicted labels and the distance between the sample and the distribution centroid.The distribution characteristics of the samples can be utilized to detect adversarial cases that can potentially evade the model defense.The effectiveness of the proposed method is demonstrated through comprehensive experiments.

On analyzing space-time distribution of evaporation duct height over the global ocean

The statistical features of the evaporation duct over the global ocean were comprehensively investigated with reanalysis data sets from the National Centers for Environmental Prediction. These data sets have time and spatial resolutions of 1 h and 0.313°x0.312°, respectively. The efficiency of the analysis was evaluated by processing weather buoy data from the Pacific Ocean and measuring propagation loss in the Yellow Sea of China. The distribution features of evaporation duct height （EDH） and the related meteorological factors for different seas were analyzed. The global EDH is generally high and demonstrates a latitudinal distribution for oceans at low latitudes. The average EDH is approximately 11 m over oceans beside the equator with a latitude of less than 20°. The reasons for the formation of the global EDH features were also analyzed for different sea areas.

Simulation and Prediction on Space-Time Distribution of Water Quality in Dianchi Lake

It is very important in pollution treatment to clarify the space-time distribution of water quality in Dianchi Lake. Based on the sample data obtained from 10 observation stations every month from 2008 to 2009, it uses space-time semi-variogram and ordinary kriging method to simulate the space-time variance and distribution of water quality indictors (TN, TP, BOD, CODMn, DO, Chlorophyl-α, etc.). Because the space-time semivariogram also has a certain of ex-trapolation function. From the simulation results, the pollution is mainly concentrated at the North (Caohai) and the Southwest area (Haikou), and water pollution has a increase trend.

Space-to-Ground Quantum Key Distribution Using a Small-Sized Payload on Tiangong-2 Space Lab

Quantum technology establishes a foundation for secure communication via quantum key distribution （QKD）. In the last two decades, the rapid development of QKD makes a global quantum communication network feasible. In order to construct this network, it is economical to consider small-sized and low-cost QKD payloads, which can be assembled on satellites with different sizes, such as space stations. Here we report an experimental demonstration of space-to-ground QKD using a small-sized payload, from Tiangong-2 space lab to Nanshan ground station. The 57.9-kg payload integrates a tracking system, a QKD transmitter along with modules for synchronization, and a laser communication transmitter. In the space lab, a 50MHz vacuum＋weak decoy-state optical source is sent through a reflective telescope with an aperture of 200mm. On the ground station, a telescope with an aperture of 1200mm collects the signal photons. A stable and high-transmittance communication channel is set up with a high-precision bidirectional tracking system, a polarization compensation module, and a synchronization system. When the quantum link is successfully established, we obtain a key rate over 100bps with a communication distance up to 719km. Together with our recent development of QKD in daylight, the present demonstration paves the way towards a practical satellite-constellation-based global quantum secure network with small-sized QKD payloads.

Space-time Dynamics of Dendroctonus valens Population in China and Spatial Sampling Technique based on Its Spatial Distribution Pattern

Red turpentine beetle （RTB）, Dendroctongs valens LeConte, is a destructive forest invasive species in China, it mainly attacks Pings tabuliformis and P. bungeana. So far it has spread rapidly to the provinces of Shanxi, Hebei, Henan, Shanxi and Beijing since its first outbreak in Shanxi Province in 1998, and has caused extensive tree mortality. Space-time dynamics of D. valens population and spatial sampling technique based on its spatial distribution pattern were ana- lyzed using geostatistical methods in the pure P. tabuliforis forests and mixedwood stands which were at different damage levels. According to the spatial distribu- tion of D. valeas population, the specific spatial sampling technique was also studied, and then was compared with traditional sampling technique. The spatial sam- piing technique combined with sampling theory and the biological characteristics of D. valens population, which not only could calcnlate the error of the sampling, but also could discuss the optimal sampling number and the optimum size of plot according to different damage levels and different stand types. This helps to explain population expansion and colonization mechanism of D. valens, and to provide a good reference for adopting snitable control measures.

Was Polchinski Wrong? Colombeau Distributional Rindler Space-Time with Distributional Levi-CivitàConnection Induced Vacuum Dominance. Unruh Effect Revisited

The vacuum energy density of free scalar quantum field in a Rindler distributional space-time with distributional Levi-Cività connection is considered. It has been widely believed that, except in very extreme situations, the influence of acceleration on quantum fields should amount to just small, sub-dominant contributions. Here we argue that this belief is wrong by showing that in a Rindler distributional background space-time with distributional Levi-Cività connection the vacuum energy of free quantum fields is forced, by the very same background distributional space-time such a Rindler distributional background space-time, to become dominant over any classical energy density component. This semiclassical gravity effect finds its roots in the singular behavior of quantum fields on a Rindler distributional space-times with distributional Levi-Cività connection. In particular we obtain that the vacuum fluctuations have a singular behavior at a Rindler horizon . Therefore sufficiently strongly accelerated observer burns up near the Rindler horizon. Thus Polchinski’s account doesn’t violate the Einstein equivalence principle.

Debris cloud structure and hazardous fragments distribution under hypervelocity yaw impact

This study investigates how the debris cloud structure and hazardous fragment distribution vary with attack angle by simulating a circular cylinder projectile hypervelocity impinging on a thin plate using the finite element-smoothed particle hydrodynamics(FE-SPH)adaptive method.Based on the comparison and analysis of the experimental and simulation results,the FE-SPH adaptive method was applied to address the hypervelocity yaw impact problem,and the variation law of the debris cloud structure with the attack angle was obtained.The screening criterion of the hazardous fragment at yaw impact is given by analyzing the debris formation obtained by the FE-SPH adaptive method,and the distribution characteristics of hazardous fragments and their relationship with the attack angle are given.Moreover,the velocity space was used to evaluate the distribution range and damage capability of asymmetric hazardous fragments.The maximum velocity angle was extended from fully symmetrical working conditions to asymmetrical cases to describe the asymmetrical debris cloud distribution range.In this range,the energy density was calculated to quantitatively analyze how much damage hazardous fragments inflict on the rear plate.The results showed that the number of hazardous fragments generated by the case near the 35°attack angle was the largest,the distribution range was the smallest,and the energy density was the largest.These results suggest that in this case,debris cloud generated by the impact had the strongest damage to the rear plate.

Analysis of temporal and spatial distribution characteristics of ammonium chloride smoke particles in confined spaces

In response to the demand for short-range detection of anti-smoke environment interference by laser fuzes,this study proposes a smoke environment simulation of non-uniform continuous point source diffusion and investigates an experimental laboratory smoke environment using an ammonium chloride smoke agent.The particle size distribution,composition,and mass flow distribution of the smoke were studied.Based on a discrete phase model and a kεturbulence model,a numerical simulation was developed to model the smoke generation and diffusion processes of the smoke agent in a confined space.The temporal and spatial distribution characteristics of the smoke mass concentration,velocity,and temperature in the space after smoke generation were analyzed,and the motion law governing the smoke diffusion throughout the entire space was summarized.Combined with the experimental verification of the smoke environment laboratory,the results showed that the smoke plume changed from fan-shaped to umbrella-shaped during smoke generation,and then continued to spread around.Meanwhile,the mass concentration of smoke in the space decreased from the middle outward;the changes in temperature and velocity were small and stable.In the diffusion stage(after 900 s),the mass concentration of smoke above 0.8 m was relatively uniform across an area of smoke that was 12 m thick.The concentration decreased over time,following a consistent decreasing trend,and the attenuation was negligible in a very short time.Therefore,this system was suitable for conducting experimental research on laser fuzes in a smoke environment.Owing to the stability of the equipment and facilities,the setup could reproduce the same experimental smoke environment by artificially controlling the smoke emission of the smoke agent.Overall,this work provides a theoretical reference for subsequent research efforts regarding the construction of uniform smoke environments and evaluating laser transmission characteristics in smoky environments.

Stress redistribution in multi-stage hydraulic fracturing of horizontal wells in shales

Multi-stage hydraulic fracturing of horizontal wells is the main stimulation method in recovering gas from tight shale gas reservoirs, and stage spacing deter- mination is one of the key issues in fracturing design. The initiation and propagation of hydraulic fractures will cause stress redistribution and may activate natural fractures in the reservoir. Due to the limitation of the analytical method in calculation of induced stresses, we propose a numerical method, which incorporates the interaction of hydraulic fractures and the wellbore, and analyzes the stress distri- bution in the reservoir under different stage spacing. Simulation results indicate the following： （1） The induced stress was overestimated from the analytical method because it did not take into account the interaction between hydraulic fractures and the horizontal wellbore. （2） The hydraulic fracture had a considerable effect on the redis- tribution of stresses in the direction of the horizontal wellbore in the reservoir. The stress in the direction per- pendicular to the horizontal wellbore after hydraulic frac- turing had a minor change compared with the original in situ stress. （3） Stress interferences among fractures were greatly connected with the stage spacing and the distance from the wellbore. When the fracture length was 200 m, and the stage spacing was 50 m, the stress redistribution due to stage fracturing may divert the original stress pat- tern, which might activate natural fractures so as to generate a complex fracture network.

Prediction of Short-Term Distributions of Load Extremes of Offshore Wind Turbines

This paper proposes a new methodology to select an optimal threshold level to be used in the peak over threshold （POT） method for the prediction of short-term distributions of load extremes of offshore wind turbines. Such an optimal threshold level is found based on the estimation of the variance-to-mean ratio for the occurrence of peak values, which characterizes the Poisson assumption. A generalized Pareto distribution is then fitted to the extracted peaks over the optimal threshold level and the distribution parameters are estimated by the method of the maximum spacing estimation. This methodology is applied to estimate the short-term distributions of load extremes of the blade bending moment and the tower base bending moment at the mudline of a monopile-supported 5MW offshore wind turbine as an example. The accuracy of the POT method using the optimal threshold level is shown to be better, in terms of the distribution fitting, than that of the POT methods using empirical threshold levels. The comparisons among the short-term extreme response values predicted by using the POT method with the optimal threshold levels and with the empirical threshold levels and by using direct simulation results further substantiate the validity of the proposed new methodology.

A Method for Improving Power Distribution Characteristics of Space Time Block Codes

Improving power distribution characteristics of space time block codes(STBCs),namely peak to average power ratio(PAPR),average to minimum power ratio(Ave/min),and probability of transmitting"zero"by antenna,makes easier their practical implementation.To this end,this study proposes to multiply full diversity STB C with a non-singular matrix in multiple input multiple output(MIMO)or multiple input single output(MISO)systems with linear or maximum likelihood(ML)receivers.It is proved that the obtained code achieves full diversity and the order of detection complexity does not change.The proposed method is applied to different types of STBCs.The bit error rate(BER)and power distribution characteristics of the new codes demonstrate the superiority of the introduced method.Further,lower and upper bounds on the BER of the obtained STBCs are derived for all receivers.The proposed method provides trade-off among PAPR,spectral efficiency,energy efficiency,and BER.

STATIONARY DISTRIBUTIONS OF INFINITE PARTICLE SYSTEMS WITH NONCOMPACT STATE SPACE.

This paper deals with the preblem of existence and uniqueness of the stationary distributions (abbr., s. d.'s) for the processes constructed in [4] .The main results are stated in § 1. For the reader's convenience we first restate the existence theorems (Theorem 1 and 2) of the processes given in [4]. Then two existence theorems (Theorem 3 and 4) and a uniqueness theorem (Theorem 5) for the s. d.'s of the processes are presented. The last result (Theorem 6), as an application of the previous ones, is about the Schlgl model which comes from nonequilibrium statisticali physics. The details of the proofs of Theorem 3—6 are given in § 2—4.

Electron beam modeling and analyses of the electric field distribution and space charge effect

In electron beam technology, one of the critical focuses of research and development efforts is on improving the measurement of electron beam parameters. The parameters are closely related to the generation, emission, operation environment, and role of the electron beam and the corresponding medium. In this study, a field calculation method is proposed, and the electric field intensity distribution on the electron beam’s cross-section is analyzed. The characteristics of beam diffusion caused by the space charge effect are investigated in simulation, and the obtained data are compared with the experiment. The simulation demonstrated that the cross-sectional electric field distribution is primarily affected by the electron beam current, current density distribution, and electron beam propagation speed.

Combination and distribution of reservoir space in complex carbonate rocks

This paper discusses the reservoir space in carbonate rocks in terms of types,combination features,distribution regularity,and controlling factors,based on core observations and tests of the North Truva Oilfield,Caspian Basin.According to the reservoir space combinations,carbonate reservoirs can be divided into four types,i.e.,pore,fracture-pore,pore-cavity-fracture,and pore-cavity.Formation and distribution of these reservoirs is strongly controlled by deposition,diagenesis,and tectonism.In evaporated platform and restricted platform facies,the reservoirs are predominately affected by meteoric fresh water leaching in the supergene-para-syngenetic period and by uplifting and erosion in the late stage,making both platform facies contain all the above-mentioned four types of reservoirs,with various pores,such as dissolved cavities and dissolved fractures,or structural fractures occasionally in favorable structural locations.In open platform facies,the reservoirs deposited continuously in deeper water,in an environment of alternative high-energy shoals（where pore-fracture-type reservoirs are dominant） and low-energy shoals（where pore reservoirs are dominant）.

Product Spacing of Stress–Strength under Progressive Hybrid Censored for Exponentiated-Gumbel Distribution

Maximum product spacing for stress–strength model based on progressive Type-II hybrid censored samples with different cases has been obtained.This paper deals with estimation of the stress strength reliability model R=P(Y<X)when the stress and strength are two independent exponentiated Gumbel distribution random variables with different shape parameters but having the same scale parameter.The stress–strength reliability model is estimated under progressive Type-II hybrid censoring samples.Two progressive Type-II hybrid censoring schemes were used,Case I:A sample size of stress is the equal sample size of strength,and same time of hybrid censoring,the product of spacing function under progressive Type-II hybrid censoring schemes.Case II:The sample size of stress is a different sample size of strength,in which the life-testing experiment with a progressive censoring scheme is terminated at a random time T 2 e0;1T.The maximum likelihood estimation and maximum product spacing estimation methods under progressive Type-II hybrid censored samples for the stress strength model have been discussed.A comparison study with classical methods as the maximum likelihood estimation method is discussed.Furthermore,to compare the performance of various cases,Markov chain Monte Carlo simulation is conducted by using iterative procedures as Newton Raphson or conjugate-gradient procedures.Finally,two real datasets are analyzed for illustrative purposes,first data for the breaking strengths of jute fiber,and the second data for the waiting times before the service of the customers of two banks.

A New Rayleigh Distribution:Properties and Estimation Based on Progressive Type-II Censored Data with an Application

In this paper,we propose a new extension of the traditional Rayleigh distribution called the modified Kies Rayleigh distribution.The new distribution contains one scale and one shape parameter and its hazard rate function can be increasing and bathtub-shaped.Some mathematical properties of the new distribution are derived including quantiles and moments.The parameters of modified Kies Rayleigh distribution are estimated based on progressively Type-II censored data.For this purpose,we consider two estimation methods,namely maximum likelihood and maximum product of spacing estimation methods.To compare the efficiency of the proposed estimators,a simulation study is carried out.To show the applicability of the new model as well as the estimation methods,one real data for failure times of software is analyzed.Based on the empirical parts,we can conclude that the proposed model can be considered as a good model in the field of life testing and reliability analysis compared with other competing models.

Investigation on the factors influencing the thickness distribution of superplastic-formed components

In the superplastic sheet forming process, the uniformity of the sheet's final thickness distribution is vital for ensuring the good mechanical quality of the formed components. The influences of the component shape and the contact friction on the final thickness distribution were investigated in this work by using finite element method on a series of axisymmetric models. It was concluded that shape optimization and friction elimination are required to get uniform thickness distribution, and eventually to improve the mechanical quality of the formed components. The constitutive equation of the Ti-6A1-4V superplastic material was also determined on the basis of experimental data.

Mechanical properties of double-stranded DNA biofilm with Gaussian distribution

In microcantilever-based label-free biodetection technologies, deflection changes induced by adsorptions of double-stranded DNA （dsDNA） molecules on Au-layer surface are greatly affected by the mechanical, thermal and electrical properties of DNA biofilm. In this paper, the elastic properties of dsDNA biofilm are studied. First, the Parsegian＇s empirical potential based on a mesoscopic liq- uid crystal theory is employed to describe the interaction energy among coarse-grained DNA cylinders. Then, con- sidering a Gaussian distribution of DNA interaxial distance, the thought experiment method is used to derive an analyti- cal expression for Young＇s modulus of DNA biofilm with a stochastic packing pattern for the first time. Results show that Young＇s modulus of DNA biofilm is on the order of 10 MPa. These findings could provide a simple and effective method to evaluate the mechanical properties of soft biofilm on snbstrate.

Distributed Multicircular Circumnavigation Control for UAVs with Desired Angular Spacing

This paper addresses a multicircular circumnavigation control for UAVs with desired angular spacing around a nonstationary target.By defining a coordinated error relative to neighboring angular spacing,under the premise that target information is perfectly accessible by all nodes,a centralized circular enclosing control strategy is derived for multiple UAVs connected by an undirected graph to allow for formation behaviors concerning the moving target.Besides,to avoid the requirement of target’s states being accessible for each UAV,fixed-time distributed observers are introduced to acquire the state estimates in a fixed-time sense,and the upper boundary of settling time can be determined offline irrespective of initial properties,greatly releasing the burdensome communication traffic.Then,with the aid of fixed-time distributed observers,a distributed circular circumnavigation controller is derived to force all UAVs to collaboratively evolve along the preset circles while keeping a desired angular spacing.It is inferred from Lyapunov stability that all errors are demonstrated to be convergent.Simulations are offered to verify the utility of proposed protocol.