In recent years, artificial intelligence technology represented by deep neural network has achieved great success in many application fields , and has gradually changed people's production and lifestyle. However, its security and robustness have also attracted wide attention. The attacks and defenses of adversarial examples have become the most important research direction in this field, and the military confrontation between the two has been evolving constantly. First of all, this paper summarizes and sorts out the adversarial example attack methods. Then, from the perspective of hardware deployment, the existing adversarial example defense methods are sorted out. Finally, the paper points out some research directions worthy of attention in the future.

Information technology brings convenience to people's life but leaks out personal privacy at the same time. Federated learning is a kind of machine learning technology that can protect data privacy. Different from current machine learning methods, federated learning is usually facing with non-independent and uniformly dis- tributed data( non-IID ) and keeps data in local places. Therefore, the effectiveness of existing machine learning methods in federated learning non-ID problems decreases greatly. Aiming at the non-IID problem in federated learning, this paper makes improvement on the basis of federated average algorithm, divides MNIST dataset into non-IID slices and distributes them to all participants, calculates the EMD ( Earth Mover's Distance) of each participant's data, takes the latter quartile distance as the upper bound, and removes the participants with too large EMD to ensure the overall effect of federation. Experimental results show that the proposed method im- proves the accuracy by about 5% compared with the existing federated average algorithm, reduces the communi- cation overhead in the federated learning training process，and improves the overall efficiency. The introduction of EMD can also provide a measurement basis for measuring the contribution of each participant.

The development of information technology has promoted social progress but has brought more chal- lenges to information security. The information security scene is intricate, and security tool development and se- curity analysis are more complicated. At present, many information security tasks can no longer be accom- plished independently by individuals and often require the cooperation of various experts. To realize this kind of information security scheme, a mature and reliable cooperation mechanism is needed. In sofware engineering， crowdsourcing is a mature distributed problem solution that reduces complexity by subdividing complex tasks into simple tasks that can be outsourced to others to perform. In recent years , some researchers have proposed intro- ducing crowdsourcing mechanisms in information security to help solve some complicated security problems ， which makes us realize the feasibility of combining information security and crowdsourcing mechanism. This pa- per will discuss the application status of crowdsourcing mechanisms in information security and put forward the future development directions.

Each Android malware instance belongs to a specific family that performs a similar set of actions and shares some key characteristics. Being competent to identify Android malware families is critical to address se- curity challenges from malware and the damaging consequences. Ensemble learning is believed an improvement to solve computational intelligence problems by strategically combining decisions from multiple machine learning models. This paper, thus ，presents a study of the application of ensemble learning and its effectiveness in An- droid malware family identification/ 'classification against other single-model-based identification approaches. To conduct a fair evaluation, a prototype of ensemble learning based Android malware classification system was de- veloped for this work, where W eighted Majority Voting ( WMV ) approach is used in this prototype to determine the importance of individual models (i. e. , Support Vector Machine, k-Nearest Neighbour, ExtraTress , Multi- layer Perceptron, and Logistic Regression) to a final decision. The results of the evaluation, conducted by using publicly -accessible malware datasets (i. e.，Drebin and UpDroid) and the recent samples from GitHub reposito- ries, show that the ensemble learning approach does not always perform better than single-model learning ap- proaches. The performance of the ensemble learning based malware family classification is heavily influenced by several factors, in particular the features, the values of the parameters and the weights assigned to individual models.

With the acceleration of information process, the global society has entered the era of big data. The innovation of the internet and the rapid rise of e-commerce make the logistics industry develop to a new stage, which is at the critical juncture of survival and development. The traditional logistics operation mode is not suit- able for such a highly information age. Therefore, how to realize the automation，informatization, and intellectu- alization using big data technology has become an important way for the innovative reformation and development of the logistics industry. This paper introduces the problems faced by the logistics industry and the practical ap- plication of big data technology in the logistics industry，providing certain innovative ideas for the logistics in- dustry in the era of big data, so as to achieve breakthrough development.

WebRTC open source noise reduction algorithm is mainly used for real-time communication on PC or . mobile, but many applications need to be implemented on embedded chip. This paper uses the cost-effective RT1052 development board as the hardware platform to analyze the algorithm processing flow and pass the hard- ware floating point processing unit ( FPU) and Signal Processing Library ( CMSIS-DSP) optimize code. To tack- le the convergence problem in speech recovery in the original code, by adding a voice activity detector ( VAD)， no noise parameter update is performed in the silent segment, and the accuracy of the noise parameter is main- tained. The test results show that the WebRTC noise reduction algorithm is successfully implemented on the RT1052 chip. The code efficiency is improved by about 4.5 times and the perceptual evaluation of speech quali- ty( PESQ) score is increased by about 0. 15.

The mixture experiment is a kind of experiment whose aim is to study the proportion of mixture com- ponents. In the mixture experiment, if the response is also affected by the categorical variable, a type of com- pound mixture model containing qualitative factors should be established to fit the actual experimental design problem. In general optimum experimental design problems, good prediction properties are essential for optimi- zation. l-optimal design is an important prediction-based experimental design. The purpose of this paper is to study the I-optimal design of mixture models with a qualitative factor, to give some theorems for I-optimal of this kind of mixture model, as well as to calculate the approximate I-optimal weights of three basic mixture models with a qualitative factor. The derived conclusions at the end of this article can be applied to actual life.

In this paper, by applying high-order two-scale method，the high-order two-scale asymptotic expan- sion for the solution of electro-thermo-mechanical coupling problem is constructed. Based on the constructed high-order asymptotic solution, the homogenization constants and the homogenization solution corresponding to the weak solution are defined respectively, the second-order approximate solution is obtained and the error esti- mation is analyzed.

The reservoir has important natural ecological , economic and social value. In recent years, the tem- poral and spatial interference of human activities to the reservoir is more and more intense. Based on the multi- phase high-resolution remote sensing images from 1990 to 2015, combined with multi-source data, this study es- tablished a national multi-phase dynamic database of reservoirs, to dynamically monitor the number and spatial distribution of reservoirs in China in recent 30 years，and revealed the spatiotemporal evolution process of reser- voirs in China under the interference of high-intensity human activities. Based on the heterogeneity and correla- tion of spatial distribution of reservoirs，the temporal and spatial changes of various factors of reservoirs and their natural and social responses are discussed by using comprehensive methods. The results show that under the in- fluence of high-intensity human activities, the number and capacity of reservoirs in China have been increasing in recent 30 years, and the spatial distribution of reservoirs tends to be dispersed，showing the phenomenon of multi-core aggregation. A large number of reservoirs and water conservancy projects have created great economic and social benefits, but at the same time, due to a large number of intercepted water and sediment, they inter- fere with the process of water and sediment balance, and bring a series of challenges to the ecological environ- ment. How to achieve the balance between the economic and social benefits and the ecological and environmen- al benefits of reservoirs is the focus of the future research and of significance to the management of reservoirs and water conservancy projects.

Since the reform and opening up, the economic development of Guangdong province is rapid, and the regional characteristics of land use structure are distinct. Therefore, it is of great significance to analyze the . stage characteristics of land use structure in Guangdong province in recent 40 years. Based on the vector data of land use types in Guangdong province in the multi-temporal remote sensing monitoring database of China's land use status, the quantitative structure analysis is carried out by combining CA-Markov model, Lorenz curve and land use structure index. The results show that: the macro-control effect on the dynamic degree of single land type is obvious. The forecast shows that the industrial and mining residential land will become the second land type in the regional land use structure in 2025 ; the distribution of grassland，industrial and mining residents and unused land tend to be concentrated; the water area and cultivated land tend to disperse; the changes of land structure diversity and dominance show that the changes of urban structure in the Pearl River Delta are more complex and changeable and the structural changes of the remaining three areas are relatively gentle. Under the accumulation of economic and social activities, the original structure of land will form a dissimilated land use structure. It is of great significance to analyze the coupling relationship between human activities and land use . from the perspectives of land class stage dynamics and regional differences.