In recent years, with the emergence of models based on Transformers and MLPs, such as Vision Transformer (ViT) and MLP-Mixer, researchers have begun to explore the potential of these new architectures in visual tasks. These models have achieved significant results, showing competitive performance compared to traditional CNNs. However, the robustness of models is another key research topic nowadays. To comprehensively evaluate the robustness of these models, researchers conducted extensive experiments on datasets including MNIST, Fashion-MNIST, and Fruit. Inspired by the above analysis, researchers proposed the MC (MLP-CNN) model, a hybrid architecture that combines the advantages of visual MLPs and convolutional neural networks. Experimental results show that the MC model demonstrated superior robustness across multiple datasets under adversarial attacks. These findings provide important references for designing more robust deep learning models and point the direction for future research in the field of image processing.

Deep learning methods have been exerting their strengths in long-term time series forecasting. However, they often struggle to strike a balance between expressive power and computational efficiency. Resorting to multi-layer perceptrons (MLPs) provides a compromising solution, yet they suffer from two critical problems caused by the intrinsic point-wise mapping mode, in terms of deficient contextual dependencies and inadequate information bottleneck. Here, we propose the Coarsened Perceptron Network (CP-Net), featured by a coarsening strategy that alleviates the above problems associated with the prototype MLPs by forming information granules in place of solitary temporal points. The CP-Net utilizes primarily a two-stage framework for extracting semantic and contextual patterns, which preserves correlations over larger timespans and filters out volatile noises. This is further enhanced by a multi-scale setting, where patterns of diverse granularities are fused towards a comprehensive prediction. Based purely on convolutions of structural simplicity, CP-Net is able to maintain a linear computational complexity and low runtime, while demonstrates an improvement of 4.1 https://github.com/nannanbian/CPNet

Reinforcement learning (RL) has emerged as a promising solution for addressing traffic signal control (TSC) challenges. While most RL-based TSC systems typically employ an online approach, facilitating frequent active interaction with the environment, learning such strategies in the real world is impractical due to safety and risk concerns. To tackle these challenges, this study introduces an innovative offline data-driven approach, called DataLight. DataLight employs effective state representations and reward function by capturing vehicular speed information within the environment. It then segments roads to capture spatial information and further enhances the spatially segmented state representations with sequential modeling. The experimental results demonstrate the effectiveness of DataLight, showcasing superior performance compared to both state-of-the-art online and offline TSC methods. Additionally, DataLight exhibits robust learning capabilities concerning real-world deployment issues. The code is available at https://github.com/LiangZhang1996/DataLight.

Recent years have witnessed a growing interest in using machine learning to predict and identify phase transitions in various systems. Here we adopt convolutional neural networks (CNNs) to study the phase transitions of Vicsek model, solving the problem that traditional order parameters are insufficiently able to do. Within the large-scale simulations, there are four phases, and we confirm that all the phase transitions between two neighboring phases are first-order. We have successfully classified the phase by using CNNs with a high accuracy and identified the phase transition points, while traditional approaches using various order parameters fail to obtain. These results indicate that the great potential of machine learning approach in understanding the complexities in collective behaviors, and in related complex systems in general.

Land use of irrigated areas nearby the metropolitan is complex. On fields, crop growth may differ with variations in the water demand. Among the image classification methods, combined object-oriented classification is currently preferred over conventional pixel-based classification. Compared with the traditional pixel-based method, which generally exhibits a spot-like salt-and-pepper effect, object-based classification can significantly reduce the salt-and-pepper effect and amount of data required for analysis. To obtain improved spectral recognition, maximum image information is described using color, texture, and shape to enhance image recognition. In this study, image information extraction and crop interpretation were performed using the airborne digital sensor ADS40 to obtain experimental data, and the traditional supervised image and image object classification methods were compared. The results indicate that both the image classification methods could yield an overall accuracy of more than 80%, and the accuracy of object-based classification (88.68%) was higher than that of the other classification. The daily water requirement of crops in the study area, calculated using a high-precision image object classification method, was approximately 2585 m(3). The current results may aid in the effective estimation of agricultural irrigation water consumption.

This study combined statistical analysis and remote sensing techniques to explore the environmental effect of land use and land cover changes in Taipei City, Taiwan. Together with SPOT satellite images from multiple periods (1993, 2003, and 2014), unsupervised Iterative Self-Organizing Data Analysis techniques were used to classify images into three categories: water, vegetation (green area), and non-planting (buildings) areas. An accuracy assessment was conducted to analyze the changes in each administrative region and to explore the differences in the green cover ratios of the various administrative regions through the spatial distribution characteristics of satellite images. We found that if the area of green cover must be increased, the need for green coverage and distribution can be quantified through geospatial analysis to identify preferred sites. In addition to increasing the proportion of green coverage in the city, this approach can effectively mitigate changes in the ambient temperature. The overall accuracy and kappa values of this study were more than 90% and 0.8, respectively, indicating that the image classification results had favorable reliability.

This study presented a design of experiences approach to solve this problem, whose steps include: 1 generation of experimental design 2 implementation of experimental design 3 construction of response variable model 4 definition of optimization problem 5 solution of optimization problem. The above step 1 to 3 is to create a model of response variables to be as an alternative for structural analysis software, and because the model is a set of regular and simple functions, it can easily define the optimization problem in step 4, and then the optimization problem can be solved with optimization software in step 5. The reason that neural network is employed instead of the traditional regression analysis in step 3 is in structures the relations between internal forces and displacements and section size of members are often nonlinear. The greatest advantage of neural networks is that it is a nonlinear system; hence, it can very precisely build a nonlinear model. In this paper, the optimization of cross section of compressive steel column is employed as the case studies to assess the feasibility of the approach. The results show that this approach can indeed get a more economical design.

This paper proposes genetic algorithm combining operation tree (GAOT) and applies it to estimate the bacillariophyta algae of Techi reservoir using Landsat 8 data. GAOT is a data mining method, used to automatically discover the relationships among nonlinear systems. The main advantage of GAOT is to optimize appropriate types of function and their associated coefficients simultaneously. In the case study, this GAOT described above combining with Landsat 8 seven bands was employed. These results are then verified with in situ algae data of Techi reservoir. The results show that the GAOT generates accurate equation and has better performance than linear regression method.

This paper is aimed at demonstrating a genetic algorithm method and applying it to predict the water quality of reservoir in Taiwan island using remote sensing data.Genetic algorithms will be combined with operation tree (GAOT) to find the relationships between input and output data.A fittest function type will be obtained automatically from this method.The advantages of GA are global optimization, nonlinearity, flexibility and parallelism.In the current case study, GA is used to construct the relationship between algae concentration and Landsat sensor data.The results show that the model has better performance than the traditional LN transform of linear regression method, and similar performance compared with back-propagation neural network (BPNN) method.

This paper represents the first study to simulate the biodegradation rates (BDRs) of several components in the sewer using the sewer biodegradation model (SBDM). In order to verify the fitness between the experimental values and model values, experiments were conducted in a 21 m long sewer pilot plant and the results showed high fitness (All correlational coefficient values were greater than 0.91). Since the SBDM was validated, the biodegradation rate (BDR) was simulated. The production of hydrogen sulfide was also simulated. The results revealed that aerobic growth of heterotrophs in biofilm predominated the biodegradation. The growth rate of heterotrophs in biofilm was greater than the decay rate from initial time to the 3rd hour, but it changed after the 4th hour because of low substrate concentration. During the experimental period, the BDRs were greater than the supply rate for five components. For two components, the supply rates were greater than the BDRs. The BDR of dissolved oxygen was greater than the supply rate before the 3rd hour but that reversed after the 3rd hour. According to the results, the sewer biodeterioration could be simulated using SBDM. (C) 2017 Elsevier Ltd. All rights reserved.

The purpose of this study is to demonstrate the use of an improved genetic algorithm combining operation tree method (IGAOT) and apply it to monitor the salinity of the Taiwan Strait by using remote-sensing data. The genetic algorithm combining operation tree (GAOT) is a data mining method used to automatically discover relationships among nonlinear systems. Based on genetic algorithms (GAs), the relationships between input and output can be expressed as parse trees. The GAOT method typically has the disadvantages of premature convergence, which means it cannot produce satisfying solutions and performs satisfactorily when applied to only low-dimensional problems. Therefore, the GAOT method is enhanced using an automatic incremental procedure to improve the search ability of the method and avoid trapping in a local optimum. In this case study, an IGAOT is used to determine the relationship between the in situ data on the salinity of the Taiwan Strait and the data on the spectral parameters, seven wavebands, of a Moderate-Resolution Imaging Spectroradiometer (MODIS) sensor. The results indicate that the IGAOT model performs more favorably than do the GAOT and linear regression (LR1 and LR2) models, exhibits higher correlation coefficients, and involves fewer estimating errors. The results of this study indicate that the proposed technique is useful for estimating the Taiwan Strait salinity.

Based on the elasto-hydrodynamic lubrication theory, a 2-degree-of-freedom nonlinear dynamic model of helical gears with double-sided film is proposed, in which the minimum film thickness behaves as a function of load parameters, lubricant parameters, and the geometry of the contact. Then, the comparison of the hysteresis loops in different gear models shows the soundness of the presented model. Using numerical method, the time evolution of lubricant normal force, minimum film thickness, and lubricant stiffness is obtained in order to demonstrate the influence of the driving torque and pinion’s velocity. The results obtained in this article can contribute to the root cause for the gear vibration and show that the hydrodynamic flank friction has almost no influence on the gear system.

This paper will analyze and explore when a self-anchored suspension bridge's main cables, main girders, towers, hangers, and other structural elements' material elastic modulus is individually altered, what the effect on the bridge's vertical earthquake responses would be in order to clearly understand the self-anchored suspension bridge's dynamic characteristics. The results of this research indicate that the effect of a change in the main cable's elastic modulus on the maximum internal forces and displacements produced by an earthquake in the vertical direction is the largest, followed by the effect of a change in the elastic modulus of the main girder, towers, and hangers.

为了查明张集矿A组煤层底板灰岩含水层突水通道及含水层水力联系,向奥陶系灰岩含水层投放NH4Cl、KI 2种示踪剂,同时向寒武系灰岩含水层投放食用胭脂红示踪剂,并在在突水点处按一定间隔采集样品,分析离子浓度及分光度随时间变化关系曲线,研究表明:奥陶系灰岩含水层中存在4条岩溶通道,寒武系灰岩含水层存在3条岩溶通道,且不同含水层水力存在一定的水力联系,突水水源为多种水源混合.

Genetic Algorithm Combining Operation Tree (GAOT) was constructed to estimate the slump flow of high-performance concrete (HPC) by using seven concrete ingredients. HPC is a highly complex material; because modeling its behavior is extremely difficult, robust optimization techniques are required. GAOT is a type of evolutionary algorithm that simultaneously optimizes functions and their associated coefficients and is suitable for automatically discovering relationships between nonlinear systems. In a case study, it was observed that for estimating HPC slump flow, the GAOT is more accurate than regression model and back-propagation neural networks (BPNN).

Epidemics often exhibit drastic dynamics, unmatched by percolation theory—a difference that may be due to cooperation between contagions. A mechanistic model implicates network topology in regulating the efficiency of this cooperation.

This paper represents the first study to compare seven types of first–order and one–variable grey differential equation model [abbreviated as GM (1, 1)] and back-propagation artificial neural network (BPNN) for predicting hourly particulate matter (PM) including PMio and PM2.5 concentrations in Dali area of Taichung City, Taiwan. Their prediction performance was also compared. The results indicated that the minimum mean absolute percentage error (MAPE), mean squared error (MSE), and root mean squared error (RMSE) was 16.76%, 132.95, and 11.53, respectively for PM10 prediction. For PM2.5 prediction, the minimum MAPE, MSE, and RMSE value of 21.64%, 40.41, and 6.36, respectively could be achieved. All statistical values revealed that the predicting performance of GM (1, 1, x(0)), GM (1, 1, a), and GM (1, 1, b) outperformed other GM (1, 1) models. According to the results, it revealed that GM (1, 1) could predict the hourly PM variation precisely even comparing with BPNN.

Rainfall is a fundamental process in the hydrologic cycle. This study investigated the cause-effect relationship in which precipitation at lower frequencies affects the amount of emitted radiation and at higher frequencies affects the amount of backscattered terrestrial radiation. Because the advantage of a probabilistic graphical model is its graphical representation, which allows easy causality interpretation using the arc directions, two Bayesian networks (BNs) were used, namely, a naive Bayes classifier and a tree-augmented naive Bayes model. To empirically evaluate and compare BN-based models, "black box"-based models, including nearest-neighbor searches and artificial neural network (ANN)-based multilayer perceptron and logistic regression, were used as benchmarks. For the two study regions-namely, the Tanshui River basin in northern Taiwan and Chianan Plain in southern Taiwan-rain occurrences during typhoon seasons were examined using passive microwave imagery recorded using the Special Sensor Microwave Imager/Sounder. The results show that although black box models exhibit excellent prediction ability, interpretation of their behavior is unsatisfactory. By contrast, probabilistic graphical models can explicitly reveal the causal relationship between brightness temperatures and nonrain/rain discrimination. For the Tanshui River basin, 19.35-, 22.23-, 37.0-, and 85.5-GHz vertically polarized brightness temperatures were found to diagnose rain occurrences. For the Chianan Plain, a more sensitive indicator of rain-scattering signals was obtained using 85-GHz measurements. The results demonstrate the potential use of BNs in identifying rain occurrences in regions with land features comprising various absorbing and scattering materials.

A two-run genetic programming (GP) is proposed to estimate the slump flow of high-performance concrete (HPC) using several significant concrete ingredients in this study. GP optimizes functions and their associated coefficients simultaneously and is suitable to automatically discover relationships between nonlinear systems. Basic-GP usually suffers from premature convergence, which cannot acquire satisfying solutions and show satisfied performance only on low dimensional problems. Therefore it was improved by an automatically incremental procedure to improve the search ability and avoid local optimum. The results demonstrated that two-run GP generates an accurate formula through and has 7.5 % improvement on root mean squared error (RMSE) for predicting the slump flow of HPC than Basic-GP.

Parallel hyper-cubic gene expression programming (GEP) was constructed to estimate the slump flow of high-performance concrete (HPC) by using seven concrete ingredients. HPC is a highly complex material; because modeling its behavior is extremely difficult, robust optimization techniques are required. Because of complications caused by high dimensionality, obtaining globally optimal or nearly optimal solutions to such problems is extremely difficult. GEP is a type of evolutionary algorithm that simultaneously optimizes functions and their associated coefficients and is suitable for automatically discovering relationships between nonlinear systems. However, basic GEP, which generally suffers from premature convergence, is often trapped in local optima. Thus, in this study, various parallel subpopulations were processed to enhance search diversity and avoid local optima during GEP optimization procedures. The hyper-cubic topology rapidly spreads excellent solutions throughout all subpopulations. In addition, a migration mechanism, which exchanges chromosomes among the subpopulations, exchanges information during joint optimization to maintain diversity. In a case study, it was observed that for estimating HPC slump flow, the parallel hyper-cubic GEP is more accurate than the basic GEP and two types of regression model. Although both back-propagation neural networks (BPNN) and the proposed methods performed similarly, the proposed methods were preferable because formulas with measurable parameters were clearly provided.