Backpropagation Gradient Descent Research Articles

Comparative Analysis of Gradient Descent Learning Algorithms in Artificial Neural Networks for Forecasting Indonesian Rice Prices

Artificial Neural Networks (ANN) are a field of computer science that mimics the way the human brain processes data. ANNs can be used to classify, estimate, predict, or simulate new data from similar sources. The commonly used algorithm for prediction in ANN is Backpropagation, which yields high accuracy but tends to be slow during the training process and is prone to local minima. To address these issues, appropriate parameters are needed in the Backpropagation training process, such as an optimal learning function. The aim of this study is to evaluate and compare various learning functions within the Backpropagation algorithm to determine the best one for prediction cases. The learning functions evaluated include Gradient Descent Backpropagation (traingd), Gradient Descent with Adaptive Learning Rate (traingda), and Gradient Descent with Momentum and Adaptive Learning Rate (traingdx). The dataset used is the average wholesale rice price in Indonesia, obtained from the Central Statistics Agency (BPS) website. The evaluation results show that the traingdx learning function with a 5-5-1 architecture model achieves the highest accuracy of 83.33%, representing an 8.3% improvement over the traingd and traingda learning functions, which both achieved a maximum accuracy of 75%. Based on this study, it can be concluded that using various learning functions in Backpropagation yields better accuracy compared to standard Backpropagation.

Interval Type-2 Mutual Subsethood Cauchy Fuzzy Neural Inference System (IT2MSCFuNIS)

An interval type-2 (IT2) mutual subsethood Cauchy fuzzy neural inference system has been proposed in this paper. The network architecture consists of 3-layers with all connection weights being IT2 Cauchy fuzzy membership functions (CMFs). The crisp inputs to the system are fuzzified into IT2CMFs with fixed centers and uncertain spreads. The hidden layer represents the rule-based knowledge. The firing degree of the antecedent part of each rule at the hidden layer is computed by aggregating the product of the mutual subsethood similarity measures between the inputs and the connection weights. A volume defuzzification is used to compute the numeric output. A gradient descent back-propagation algorithm is used to train the model. The novelty of the proposed model is threefold. First, is enriching the theory of the mutual subsethood fuzzy neural models by adopting the Cauchy membership function (CMF) as another powerful fuzzy basis function (FBF) rather than the classical choice of Gaussian fuzzy membership functions (GMFs). Second, is the success of computing the mutual subsethood similarity measure between the IT2CMFs and all the model parameters’ updating equations in analytic closed-form formulas, not numerically or approximately. Third, is the ability to extract the type-1 (T1) mutual subsethood Cauchy fuzzy neural inference system (T1MSCFuNIS) with all its analytic closed-form formulas directly as a special case from the general formulas of IT2MSCFuNIS model. Such a novelty makes the proposed model a concrete and effective development of the theory of mutual subsethood fuzzy neural models. Both IT2MSCFuNIS and T1MSCFuNIS models have been tested using different examples from the domains of function approximation, classification, and prediction. The results ensure the efficacy of both models compared with other models reported in the literature.

Hybrid Naïve Back Propagation Based iCOSϕ Algorithm for Enriching the Grid Performance with Photovoltaics/Battery/Ultracapacitor/Fuel Cell

This paper presents a novel method for the combination of photovoltaic (PV), fuel cell (FC), battery, and ultracapacitor systems with D-STATCOM (Distribution Static Compensator) for grid integration. In this study, a novel naive backpropagation algorithm (NBP) based iCOSϕ is proposed for obtaining fundamental components from load current for effective harmonics compensation and contributes power quality enhancement by delivering power to the grid and connected loads. Further, a modified incremental conductance (MIC) approach is employed to acquire maximum power in the event of different atmospheric conditions. The performance of the proposed method is investigated under four uncertain conditions such as (a) linear loads, (b) zero voltage regulation under dynamic loads, (c) non-linear loads, and (d) variations in solar irradiance. Moreover, the breakthrough of the developed method is compared with various existing techniques like synchronous reference frame (SRF) theory, iCosϕ, fuzzy logic controller (FLC), conductance fryze, adaptive neuro-fuzzy inference system, and gradient descent back propagation learning (GDBP) neural network (NN) based iCosϕ. The simulation outcomes disclosed that the proposed NBP based iCOSϕ technique rendered a prolific performance rather than other existing methods under all uncertain conditions.

Human Walking Gait Classification Utilizing an Artificial Neural Network for the Ergonomics Study of Lower Limb Prosthetics

Prosthetics and orthotics research, studies, and technologies have been evolving through the years. According to World Health Organization (WHO) data, it is estimated that, globally, 35–40 million people require prosthetics and orthotics usage in daily life. Prosthetics and orthotics demand is increasing due to certain factors. One of the factors is vascular-related disease, which leads to amputation. Prosthetic usage can increase an amputee’s quality of life. Therefore, studies of the ergonomic design of prosthetics are important. The ergonomic factor in design delivers prosthetic products that are comfortable for daily use. One way to incorporate the ergonomic design of prosthetics is by studying the human walking gait. This paper presents a multiclassification of human walking gait based on electromyography (EMG) signals using a machine learning method. An EMG sensor was attached to the bicep femoris longus and gastrocnemius lateral head to acquire the EMG signal. The experiment was conducted by volunteers during normal walking activity at various speeds and the movements were segmented as initial contact, which was labeled as initial gait; loading response to the terminal stance, which was labeled as mid-gait; and pre-swing to terminal swing, which was labeled as final gait. The EMG signal was then characterized using an artificial neural network (ANN) and compared to six training accuracy methods, i.e., the Levenberg–Marquardt backpropagation training algorithm, quasi-Newton training method, Bayesian regulation backpropagation training method, gradient descent backpropagation, gradient descent with adaptive learning rate backpropagation, and one-step secant backpropagation. The machine learning study performed well in the classification of three classes of human walking gait with an overall accuracy (training, testing, and validation) of 96% for Levenberg–Marquardt backpropagation. The gait data will be used to explore the design of lower limb prosthetics in future research.

Comparison of Different Parameters of Feedforward Backpropagation Neural Networks in DEM Height Estimation for Different Terrain Types and Point Distributions

Digital Elevation Models (DEMs) are commonly used for environment, engineering, and architecture-related studies. One of the most important factors for the accuracy of DEM generation is the process of spatial interpolation, which is used for estimating the height values of the grid cells. The use of machine learning methods, such as artificial neural networks for spatial interpolation, contributes to spatial interpolation with more accuracy. In this study, the performances of FBNN interpolation based on different parameters such as the number of hidden layers and neurons, epoch number, processing time, and training functions (gradient optimization algorithms) were compared, and the differences were evaluated statistically using an analysis of variance (ANOVA) test. This research offers significant insights into the optimization of neural network gradients, with a particular focus on spatial interpolation. The accuracy of the Levenberg–Marquardt training function was the best, whereas the most significantly different training functions, gradient descent backpropagation and gradient descent with momentum and adaptive learning rule backpropagation, were the worst. Thus, this study contributes to the investigation of parameter selection of ANN for spatial interpolation in DEM height estimation for different terrain types and point distributions.

Machine learning algorithms for high-resolution prediction of spatiotemporal distribution of air pollution from meteorological and soil parameters

This study uses machine learning (ML) models for a high-resolution prediction (0.1°×0.1°) of air fine particular matter (PM2.5) concentration, the most harmful to human health, from meteorological and soil data. Iraq was considered the study area to implement the method. Different lags and the changing patterns of four European Reanalysis (ERA5) meteorological variables, rainfall, mean temperature, wind speed and relative humidity, and one soil parameter, the soil moisture, were used to select the suitable set of predictors using a non-greedy algorithm known as simulated annealing (SA). The selected predictors were used to simulate the temporal and spatial variability of air PM2.5 concentration over Iraq during the early summer (May-July), the most polluted months, using three advanced ML models, extremely randomized trees (ERT), stochastic gradient descent backpropagation (SGD-BP) and long short-term memory (LSTM) integrated with Bayesian optimizer. The spatial distribution of the annual average PM2.5 revealed the population of the whole of Iraq is exposed to a pollution level above the standard limit. The changes in temperature and soil moisture and the mean wind speed and humidity of the month before the early summer can predict the temporal and spatial variability of PM2.5 over Iraq during May-July. Results revealed the higher performance of LSTM with normalized root-mean-square error and Kling-Gupta efficiency of 13.4% and 0.89, compared to 16.02% and 0.81 for SDG-BP and 17.9% and 0.74 for ERT. The LSTM could also reconstruct the observed spatial distribution of PM2.5 with MapCurve and Cramer's V values of 0.95 and 0.91, compared to 0.9 and 0.86 for SGD-BP and 0.83 and 0.76 for ERT. The study provided a methodology for forecasting spatial variability of PM2.5 concentration at high resolution during the peak pollution months from freely available data, which can be replicated in other regions for generating high-resolution PM2.5 forecasting maps.

Obstructive Sleep Apnea Detection using Frequency Analysis of Electrocardiographic RR Interval and Machine Learning Algorithms.

Obstructive Sleep Apnea (OSA) is a respiratory disorder due to obstructive upper airway (mainly in the oropharynx) periodically during sleep. The common examination used to diagnose sleep disorders is Polysomnography (PSG). Diagnose with PSG feels uncomfortable for the patient because the patient's body is fitted with many sensors. This study aims to propose an OSA detection using the Fast Fourier Transform (FFT) statistics of electrocardiographic RR Interval (R interval from one peak to the peak of the pulse of the next pulse R) and machine learning algorithms. In this case-control study, data were taken from the Massachusetts Institute of Technology at Beth Israel Hospital (MIT-BIH) based on the Apnea ECG database (RR Interval). The machine learning algorithms were Linear Discriminant Analysis (LDA), Artificial Neural Network (ANN), K-Nearest Neighbors (K-NN), and Support Vector Machine (SVM). The OSA detection technique was designed and tested, and five features of the FFT were examined, namely mean (f1), Shannon entropy (f2), standard deviation (f3), median (f4), and geometric mean (f5). The OSA detection found the highest performance using ANN. Among the ANN types tested, the ANN with gradient descent backpropagation resulted in the best performance with accuracy, sensitivity, and specificity of 84.64%, 94.21%, and 64.03%, respectively. The lowest performance was found when LDA was applied. ANN with gradient-descent backpropagation performed higher than LDA, SVM, and KNN for OSA detection.

Adaptive-neuro-fuzzy-inference-system model for prediction of ultimate load of rectangular concrete-filled steel tubular columns

This study is devoted to the development of an Adaptive-Neuro-Fuzzy-Inference-System (ANFIS) model for the prediction of ultimate load of rectangular concrete-filled steel tubular structural members. The learning process of the model is performed by conducting a combination of backpropagation gradient descent and least-squares techniques. The performance of the model is examined by several quality metrics such as coefficient of determination (R2), Root-Mean-Squared-Error (RMSE), Mean-Absolute-Error (MAE), Index of Agreement (IA) and Slope of linear regression. Monte Carlo random sampling technique is employed to propagate input variations to the output response. Moreover, the performance of ANFIS is also compared with other machine learning models including Artificial Neural Network (ANN), Support Vector Machine (SVM), Gaussian Process Regression (GPR) and Ensemble. Results show that the ANFIS model yields higher prediction performance than other machine learning models, for both training and testing data points and regarding all quality metrics. For instance, using training data points, the ANFIS model exhibits a RMSE of 0.0283 compared to 0.0342, 0.0588, 0.0291, and 0.0464 using ANN, Ensemble, GPR, and SVM, respectively (the corresponding gain values are+17.3%,+51.9%,+2.8%, and+39.0%, respectively). On the other hand, using testing data points, the ANFIS model exhibits a RMSE of 0.0276 compared to 0.0393, 0.0987, 0.0403, and 0.0460 using ANN, Ensemble, GPR, and SVM, respectively (the corresponding values of gain are+29.8%,+72.1%,+31.5%, and+40.0%, respectively). The same observation can be made for other quality metrics. It can be concluded that the ANFIS model outperforms other models for both training and testing datasets. The ANFIS model is also compared with existing works in the past, showing its improvement in prediction results. Finally, sensitivity analysis is performed to determine the degree of effect of the input parameters on the ultimate load.

The feasibility of designing a back propagation neural network to predict the levitation force of high-temperature superconducting magnetic levitation

The levitation force between the superconductor and the magnet is highly nonlinear and affected by the coupling of multiple factors, which brings many obstacles to research and application. In addition to experimental methods and finite element simulations, the booming artificial neural network (ANN) which is adept at continuous nonlinear fitting may provide another solution to predict the levitation force. And this topic has not been deeply investigated so far. Therefore, this study aims to apply the ANN to predict the levitation force, and a typical neural network applied with the back propagation (BP) is adopted. The data set with 2399 pieces of data considers nine input factors and one force output, which was experimentally obtained by several test devices. The pre-process of the data set contains cleaning, balancing, one-hot encoding (for the discrete classified variable), normalization (for the continuous variable) and randomization. A classical perception with three layers (input, hidden and output layer) is applied in this paper. And the gradient descent back propagation algorithm reduces the error by iteration. Through the assessment and evaluation of the network, a great prediction accuracy could achieve. The prediction results could well illustrate the features of force (nonlinear, hysteresis, external field dependence and type difference between the bulk and stack), which confirm the feasibility of using a BP neural network to predict the levitation force. Furthermore, the performance of the neural network is determined by the data set, especially the uniformity and balance among factors in the set. Moreover, the huge gap in the quantity of data between factors disturbs the network to make a comprehensive judgment, and in this situation, the binary one-hot encoding of the small quantity and discrete data factor is efficient, instead of the actual value of the factor, the one-hot encoded data only represent the category. Moreover, a label encoder method is adopted to distinguish the decent and ascend (decent = 1, ascent = 0) for the force hysteresis.

Comparative Evaluation of Predicting Energy Consumption of Absorption Heat Pump with Multilayer Shallow Neural Network Training Algorithms

The performance of various multilayer neural network algorithms to predict the energy consumption of an absorption chiller in an air conditioning system under the same conditions was compared and evaluated in this study. Each prediction model was created using 12 representative multilayer shallow neural network algorithms. As training data, about a month of actual operation data during the heating period was used, and the predictive performance of 12 algorithms according to the training size was evaluated. The prediction results indicate that the error rates using the measured values are 0.09% minimum, 5.76% maximum, and 1.94 standard deviation (SD) for the Levenberg–Marquardt backpropagation model and 0.41% minimum, 5.05% maximum, and 1.68 SD for the Bayesian regularization backpropagation model. The conjugate gradient with Polak–Ribiére updates backpropagation model yielded lower values than the other two models, with 0.31% minimum, 5.73% maximum, and 1.76 SD. Based on the results for the predictive performance evaluation index, CvRMSE, all other models (conjugate gradient with Fletcher–Reeves updates backpropagation, one-step secant backpropagation, gradient descent with momentum and adaptive learning rate backpropagation, gradient descent with momentum backpropagation) except for the gradient descent backpropagation model yielded results that satisfy ASHRAE (American Society of Heating, Refrigerating and Air-Conditioning Engineers) Guideline 14. The results of this study confirm that the prediction performance may differ for each multilayer neural network training algorithm. Therefore, selecting the appropriate model to fit the characteristics of a specific project is essential.

Artificial Neural Network Context-Aware Reasoning Engine Model for Predicting User-Ranked Activities

The choice of users’ activity in a context-aware environment depends on users’ preferences and background. Users tend to rank concurrent activities and select their preferred activity. Researchers and developers of context-aware applications have sought various mechanisms to implement context reasoning engines. Recent implementations use Artificial Neural Networks (ANN) and other machine learning techniques to develop a context-aware reasoning engine to predict users’ activities. However, the complexities of these mechanisms overwhelm the processing capabilities and storage capacity of mobile devices. The study models a context-aware reasoning engine using a multi-layered perceptron with a gradient descent back-propagation algorithm to predict activity from user-ranked activities using a stochastic learning mode with a constant learning rate. The work deduced that working with specific rules in training a neural network is not always applicable. Training a network without approximation of neuron’s output to the nearest whole number increases the accuracy level of the network at the end of the training.

Optical Recognition of Handwritten Logic Formulas Using Neural Networks

In this paper, we present a handwritten character recognition (HCR) system that aims to recognize first-order logic handwritten formulas and create editable text files of the recognized formulas. Dense feedforward neural networks (NNs) are utilized, and their performance is examined under various training conditions and methods. More specifically, after three training algorithms (backpropagation, resilient propagation and stochastic gradient descent) had been tested, we created and trained an NN with the stochastic gradient descent algorithm, optimized by the Adam update rule, which was proved to be the best, using a trainset of 16,750 handwritten image samples of 28 × 28 each and a testset of 7947 samples. The final accuracy achieved is 90.13%. The general methodology followed consists of two stages: the image processing and the NN design and training. Finally, an application has been created that implements the methodology and automatically recognizes handwritten logic formulas. An interesting feature of the application is that it allows for creating new, user-oriented training sets and parameter settings, and thus new NN models.

Novel Model to Predict Critical Strain Energy Release Rate in Semi-Circular Bend Test as Fracture Parameter for Asphalt Mixtures Using an Artificial Neural Network Approach

Growing use of recycled asphalt materials in asphalt pavement means the current volumetric-based Superpave mixture design may not address durability concerns arising from replacement of a proportion of virgin binder with recycled ones. To address this limitation, performance-based testing is introduced to supplement conventional volumetric mixture design in assessing cracking performance of asphalt mixtures. Louisiana Department of Transportation and Development’s Specifications for Roads and Bridges specify a criterion for the critical strain energy release rate, Jc, obtained from semi-circular bend (SCB) test as a complement of current practice to evaluate cracking resistance of asphalt mixtures. Quality control/assurance practices, however, require SCB samples to be long-term aged for five days at 85°C, which is a time-consuming process. Therefore, it is beneficial to be able to estimate SCB Jc for long-term aged asphalt mixtures based on SCB Jc measured from plant-produced asphalt mixtures. Asphalt mixture aging is complex, and various variables are involved in the aging process, including volumetric properties of asphalt mixture and chemical/rheological characteristics of asphalt binder. With the capability of artificial neural network (ANN) to address complex relationships between input and output variables, this study aims to predict the fracture parameter, SCB Jc, of asphalt mixtures using ANN. A total of 34 asphalt mixtures were selected for this study. SCB fracture test and asphalt binder tests for chemical and rheological characterization were conducted. Stepwise regression analysis was used to determine the significant parameters in the correlation with SCB Jc. With determined significant parameters, ANN using the gradient descent backpropagation approach was then applied to develop and validate the predictive model. It was shown that the developed ANN model was able to predict the fracture parameter, SCB Jc, of asphalt mixtures more accurately than linear and non-linear regression models.

Quality Evaluation Method of a Mathematics Teaching Model Reform Based on an Improved Genetic Algorithm

The poor comprehensiveness of the evaluation indexes of quality evaluation methods for the traditional college mathematics teaching model reform results in low accuracy of the evaluation outcomes. In this paper, aiming at this problem, a quality evaluation method for the college mathematics teaching model reform, based on the genetic algorithm, is proposed. The simulated annealing algorithm uses the weighted comprehensive objective evaluation multiobjective optimization effect that can effectively improve the accuracy of the evaluation results. In the training process, the gradient descent back-propagation training method is used to obtain new weights for the quality evaluation of college mathematics teaching mode reforms and to score various indicators and evaluate the indicators. The mean value of the outcomes is the result of mathematics teaching quality evaluation. The experimental results show that the training error of the convolutional network of the proposed method is significantly small. Based on the genetic algorithm that improves the convolutional network training process, the obtained quality evaluation outcomes are higher in accuracy, better in goodness of fitness function, and considerably lower than other state-of-the-art methods. We observed that the improved genetic algorithm has a more than 90% goodness of fit and the error is significantly lower, that is, 0.01 to 0.04, than the classical genetic algorithm.

Investigations on the relationship among the porosity, permeability and pore throat size of transition zone samples in carbonate reservoirs using multiple regression analysis, artificial neural network and adaptive neuro-fuzzy interface system

Finding an accurate method for estimating permeability aside from well logs has been a difficult task for many years. The most commonly used methods targeted towards regression technique to understand the correlation between pore throat radii, porosity and permeability are Winland and Pittman equation approaches. While these methods are very common among petrophysicists, they do not give a good prediction in certain cases. Consequently, this paper investigates the relationship among porosity, permeability, and pore throat radii using three methods such as multiple regression analysis, artificial neural network (ANN), and adaptive neuro-fuzzy inference system (ANFIS) for application in transition zone permeability modeling. Firstly, a comprehensive mercury injection capillary pressure (MICP) test was conducted using 228 transition zone carbonate core samples from a field located in the Middle-East region. Multiple regression analysis was later performed to estimate the permeability using pore throat and porosity measurement. For the ANN, a two-layer feed-forward neural network with sigmoid hidden neurons and a linear output neuron was used. The technique involves training, validation, and testing of input/output data. However, for the ANFIS method, a hybrid optimization consisting of least-square and backpropagation gradient descent methods with a subtractive clustering technique was used. The ANFIS combines both the artificial neural network and fuzzy logic inference system (FIS) for the training, validation, and testing of input/output data. The results show that the best correlation for the multiple regression technique is achieved for pore throat radii with 35% mercury saturation (R35). However, for both the ANN and ANFIS techniques, pore throat radii with 55% mercury saturation (R55) gives the best result. Both the ANN and ANFIS are later found to be more effective and efficient and thus recommended as compared with the multiple regression technique commonly used in the industry.

Exploring Optimized Spiking Neural Network Architectures for Classification Tasks on Embedded Platforms.

In recent times, the usage of modern neuromorphic hardware for brain-inspired SNNs has grown exponentially. In the context of sparse input data, they are undertaking low power consumption for event-based neuromorphic hardware, specifically in the deeper layers. However, using deep ANNs for training spiking models is still considered as a tedious task. Until recently, various ANN to SNN conversion methods in the literature have been proposed to train deep SNN models. Nevertheless, these methods require hundreds to thousands of time-steps for training and still cannot attain good SNN performance. This work proposes a customized model (VGG, ResNet) architecture to train deep convolutional spiking neural networks. In this current study, the training is carried out using deep convolutional spiking neural networks with surrogate gradient descent backpropagation in a customized layer architecture similar to deep artificial neural networks. Moreover, this work also proposes fewer time-steps for training SNNs with surrogate gradient descent. During the training with surrogate gradient descent backpropagation, overfitting problems have been encountered. To overcome these problems, this work refines the SNN based dropout technique with surrogate gradient descent. The proposed customized SNN models achieve good classification results on both private and public datasets. In this work, several experiments have been carried out on an embedded platform (NVIDIA JETSON TX2 board), where the deployment of customized SNN models has been extensively conducted. Performance validations have been carried out in terms of processing time and inference accuracy between PC and embedded platforms, showing that the proposed customized models and training techniques are feasible for achieving a better performance on various datasets such as CIFAR-10, MNIST, SVHN, and private KITTI and Korean License plate dataset.

Intuitionistic fuzzy Logic System and its Application to Global Carbon Dioxide Emissions Prediction

In this study, a Takagi-Sugeno-Kang based intuitionistic fuzzy logic system is proposed for the prediction of global carbon dioxide emission for the first time. The intuitionistic fuzzy logic system is an integration of artificial neural network learning and intuitionistic fuzzy logic reasoning. The gradient descent back propagation is applied in the optimization of the parameters of the proposed model. The model is evaluated based on some performance metrics. Results of evaluation revealed that the intuitionistic fuzzy logic system outperforms other existing models in the literature in terms of prediction accuracy

Prediction of COVID-19 Time Series – Case Studies of South Africa and Egypt using Interval Type-2 Fuzzy Logic System

COVID-19 is a virus known to emanate from Wuhan, China in December 2019. COVID-19 spread widely to nearby countries like Japan and Korea, followed by Europe and America and later to Africa. Particularly, South Africa and Egypt have been worst hit by the virus. Generally, the COVID-19 data is highly uncertain and requires fuzzy logic approaches for the effective handling of these uncertainties. This study therefore presents the prediction of COVID-19 cases in South Africa and Egypt using interval type-2 fuzzy logic system with Takagi-Sugeno-Kang fuzzy inference and neural network learning. The parameters of the model are adapted using gradient descent backpropagation approach. The proposed model is found to outperform type-1 fuzzy logic system and artificial neural network in terms of the root mean squared error, mean absolute percentage error and mean absolute error

Deep neural network-based hybrid modelling for development of the cyclist infrastructure safety model

This paper is concerned with modelling cyclist road safety by considering various factors including infrastructure, spatial, personal and environmental variables affecting cycling safety. Age is one of the personal attributes, reported to be a significant critical variable affecting safety. However, very few works in the literature deal with such a problem or undertaking modelling of this variable. In this work, we propose a hybrid approach by combining statistical and supervised deep learning with neural network classifier, and gradient descent backpropagation error function for road safety investigation. The study area of Tyne and Wear County in the north-east of England is used as a case study. An accurate dynamic road safety model is constructed, and an understanding of the key parameters affecting the cyclist safety is developed. It is hoped that this research will help in reducing the cyclist crash and contribute towards sustainable integrated cycling transportation system, by making use of cut above methodologies such as deep learning neural network.

Как обмануть нейронную сеть? Синтез шума для уменьшения точности нейросетевойклассификации изображений

The study aims to develop an algorithm and then software to synthesise noise that could be used to attack deep learning neural networks designed to classify images. We present the results of our analysis of methods for conducting this type of attacks. The synthesis of attack noise is stated as a problem of multidimensional constrained optimization. The main features of the attack noise synthesis algorithm proposed are as follows: we employ the clip function to take constraints on noise into account; we use the top-1 and top-5 classification error ratings as attack noise efficiency criteria; we train our neural networks using backpropagation and Adam's gradient descent algorithm; stochastic gradient descent is employed to solve the optimisation problem indicated above; neural network training also makes use of the augmentation technique. The software was developed in Python using the Pytorch framework to dynamically differentiate the calculation graph and runs under Ubuntu 18.04 and CentOS 7. Our IDE was Visual Studio Code. We accelerated the computation via CUDA executed on a NVIDIA Titan XP GPU. The paper presents the results of a broad computational experiment in synthesising non-universal and universal attack noise types for eight deep neural networks. We show that the attack algorithm proposed is able to increase the neural network error by eight times