NN Approach Research Articles

The Development of a Prediction Model Related to Food Loss and Waste in Consumer Segments of Agrifood Chain Using Machine Learning Methods

Food loss and waste (FLW) is a primary focus topic related to all human activity. This phenomenon has a great deal of importance due to its effect on the economic and social aspects of human systems. The most integrated approach to food waste analysis is based on the study of FLW alongside the agrifood chain, which has also been performed in previous studies by the authors. This paper presents a modality of determination of food loss and waste effects with an emphasis on consumer segments in agrifood chains in the form of a predictive model based on statistical data collected based on specific methods in Romania. The determination is made comparatively, using two predictive machine learning-based methods and separate instruments (software), in order to establish the best model that fits the collected data structure. In this matter, a Decision Tree Approach (DTA) and a Neural Network Approach (NNA) will be developed, and common methodologies of the approaches will be applied. The results will determine predictive outcomes for a specific food waste (FW) agent (e.g., consumer) based on pattern recognition of the collected data. The results showed relatively high-accuracy predictions, especially for the NN approach, with lower performances using the DTA. The effects of the application of this predictive model will be expected to improve the food loss prevention measures within economic contexts when applied to real-life scenarios.

Constructing a Precise Method to Control Non-Linear Systems Employing Special Functions and Machine Learning

This paper is about finding the best way to control certain types of equations that describe heat and diffusion using methods that are not exact but close to the best solution. A new way to control things better is suggested using patterns we've seen before and computer learning. By employing the data gathered from the EEFs are computed using the Karhunen-Loève decompose in the PDE computer. The EEFs are used to convert the PDE system into a high-order ODE system. A smaller model (ROM) that may depict the primary behavior of the PDE subsystem is created using the SP approach. This facilitates understanding of the system. In order to reduce its size, a more basic model (ROM) that demonstrates the primary motions of the PDE system is created using the SP approach. Next, the optimal controller for the ROM is designed using the HJB technique. This ensures that the high-order ODE system is stable using the SP theory. By splitting the best way to control something into two sections, we can solve a math problem to figure out the first part and come up with a new equation for the second part. Third, a method is suggested for updating and controlling based on small improvements to solve a specific equation, and it has been proven to work effectively. Moreover, we use a technique called the NN approach to estimate the cost function. Finally, the simulation results demonstrate the effectiveness of the proposed optimum control approach when applied to a diffusion-reaction process using a unique spatially component.

Adaptive stabilization of constrained stochastic nonlinear systems with input saturation: A combined BLF and NN approach

This paper investigates the adaptive control for a class of constrained stochastic nonlinear systems with parametric uncertainty and input saturation. Based on a novel radial basis function neural networks (RBF NNs), the nonlinearities are tackled without the prior knowledge of NN nodes and weights. The approximate coordinate coordination is combined with an auxiliary system to attenuate the effects generated by input saturation. Then, an opportune backstepping design procedure is presented using the barrier Lyapunov function (BLF) and RBF NN. Based on this design procedure, an adaptive state–feedback controller is constructed, which makes the closed-loop system semi-globally uniformly ultimately bounded, the tracking error bounded in a compact set of the origin, and the full-states not violated. Finally, a stochastic single-link robot arm system is simulated to demonstrate the effectiveness of the proposed scheme.

Lifelong deep learning‐based control of robot manipulators

SummaryThis study proposes a lifelong deep learning control scheme for robotic manipulators with bounded disturbances. This scheme involves the use of an online tunable deep neural network (DNN) to approximate the unknown nonlinear dynamics of the robot. The control scheme is developed by using a singular value decomposition‐based direct tracking error‐driven approach, which is utilized to derive the weight update laws for the DNN. To avoid catastrophic forgetting in multi‐task scenarios and to ensure lifelong learning (LL), a novel online LL scheme based on elastic weight consolidation is included in the DNN weight‐tuning laws. Our results demonstrate that the resulting closed‐loop system is uniformly ultimately bounded while the forgetting is reduced. To demonstrate the effectiveness of our approach, we provide simulation results comparing it with the conventional single‐layer NN approach and confirm its theoretical claims. The cumulative effect of the error and control input in the multitasking system shows a 43% improvement in performance by using the proposed LL‐based DNN control over recent literature.

Finite element and generalized regression neural network modelling of multiple cracks growth under the influence of multiple crack parameters

PurposeThis study presents finite element (FE) and generalized regression neural network (GRNN) approaches for modeling multiple crack growth problems and predicting crack-growth directions under the influence of multiple crack parameters.Design/methodology/approachTo determine the crack-growth direction in aluminum specimens, multiple crack parameters representing some degree of crack propagation complexity, including crack length, inclination angle, offset and distance, were examined. FE method models were developed for multiple crack growth simulations. To capture the complex relationships among multiple crack-growth variables, GRNN models were developed as nonlinear regression models. Six input variables and one output variable comprising 65 training and 20 test datasets were established.FindingsThe FE model could conveniently simulate the crack-growth directions. However, several multiple crack parameters could affect the simulation accuracy. The GRNN offers a reliable method for modeling the growth of multiple cracks. Using 76% of the total dataset, the NN model attained an R2 value of 0.985.Research limitations/implicationsThe models are presented for static multiple crack growth problems. No material anisotropy is observed.Practical implicationsIn practical crack-growth analyses, the NN approach provides significant benefits and savings.Originality/valueThe proposed GRNN model is simple to develop and accurate. Its performance was superior to that of other NN models. This model is also suitable for modeling multiple crack growths with arbitrary geometries. The proposed GRNN model demonstrates its prediction capability with a simpler learning process, thus producing efficient multiple crack growth predictions and assessments.

Node Classification on The Citation Network Using Graph Neural Network

Research on Graph Neural Networks has influenced various current real-world problems. The graph-based approach is considered capable of effectively representing the actual state of surrounding data by utilizing nodes, edges, and features. Consider the feedforward neural network and the graph neural network approaches, we determine the accuracy of each method. In the baseline experiment, training and testing were performed using the NN approach. The resulting accuracy of FNN was 72.59% and GNN model has increased by 81.65%. There is a 9.06% increase in accuracy between the baseline model and the GNN model. The new data utilized in the model predictions showcases the probabilities of each class through randomly generated examples.

A novel electricity load forecasting based on probabilistic least absolute shrinkage and selection operator-Quantile regression neural network

Predicting energy consumption is challenging since forecast accuracy is heavily impacted by several external elements, including the economy, community, environment, and sustainable energy, including wind, solar, and hydro energy. The most crucial aspect in successfully predicting electrical power demand is to gather crucial information about those external factors in the vast amount of data provided by the smart grid. The Least Absolute Shrinkage and Selection Operator-Quantile Regression Neural Network (LASSO-QRNN) is a strategy for probability density prediction proposed in this article. First, using LASSO regression, the primary features are determined from external factors impacting electrical power utilization. The LASSO-Quantile regression NN approach is then developed to forecast annual electrical power consumption. In the years that followed, various quantiles were employed to gauge electrical power demand, and the outcomes were assessed. The LASSO-QRNN approach also includes a kernel density evaluation that, in the event of a one-value forecast, will display a probability distribution. The experimental assessments from the Guangdong database in China and the California real-world databases in the US have evaluated the forecasting precision. The results of the numerical study demonstrate that, in comparison to Quantile Regression and Kernel Density, Box-Cox transformation, the provided model provides superior accuracy for the prediction of electrical power demand. Deep Generative Architecture, K-means Clustering, a Quantile Regression Neural Network, expert prediction, and fuzzy Bayesian theory. LASSO-Quantile regression neural network can handle the higher dimension information in electrical power usage prediction besides the more accurate outcomes.

ReLU Neural Network Galerkin BEM

We introduce Neural Network (NN for short) approximation architectures for the numerical solution of Boundary Integral Equations (BIEs for short). We exemplify the proposed NN approach for the boundary reduction of the potential problem in two spatial dimensions. We adopt a Galerkin formulation-based method, in polygonal domains with a finite number of straight sides. Trial spaces used in the Galerkin discretization of the BIEs are built by using NNs that, in turn, employ the so-called Rectified Linear Units (ReLU) as the underlying activation function. The ReLU-NNs used to approximate the solutions to the BIEs depend nonlinearly on the parameters characterizing the NNs themselves. Consequently, the computation of a numerical solution to a BIE by means of ReLU-NNs boils down to a fine tuning of these parameters, in network training. We argue that ReLU-NNs of fixed depth and with a variable width allow us to recover well-known approximation rate results for the standard Galerkin Boundary Element Method (BEM). This observation hinges on existing well-known properties concerning the regularity of the solution of the BIEs on Lipschitz, polygonal boundaries, i.e. accounting for the effect of corner singularities, and the expressive power of ReLU-NNs over different classes of functions. We prove that shallow ReLU-NNs, i.e. networks having a fixed, moderate depth but with increasing width, can achieve optimal order algebraic convergence rates. We propose novel loss functions for NN training which are obtained using computable, local residual a posteriori error estimators with ReLU-NNs for the numerical approximation of BIEs. We find that weighted residual estimators, which are reliable without further assumptions on the quasi-uniformity of the underlying mesh, can be employed for the construction of computationally efficient loss functions for ReLU-NN training. The proposed framework allows us to leverage on state-of-the-art computational deep learning technologies such as TENSORFLOW and TPUs for the numerical solution of BIEs using ReLU-NNs. Exploratory numerical experiments validate our theoretical findings and indicate the viability of the proposed ReLU-NN Galerkin BEM approach.

Anomaly Detection in Financial Time Series by Principal Component Analysis and Neural Networks

A major concern when dealing with financial time series involving a wide variety of market risk factors is the presence of anomalies. These induce a miscalibration of the models used to quantify and manage risk, resulting in potential erroneous risk measures. We propose an approach that aims to improve anomaly detection in financial time series, overcoming most of the inherent difficulties. Valuable features are extracted from the time series by compressing and reconstructing the data through principal component analysis. We then define an anomaly score using a feedforward neural network. A time series is considered to be contaminated when its anomaly score exceeds a given cutoff value. This cutoff value is not a hand-set parameter but rather is calibrated as a neural network parameter throughout the minimization of a customized loss function. The efficiency of the proposed approach compared to several well-known anomaly detection algorithms is numerically demonstrated on both synthetic and real data sets, with high and stable performance being achieved with the PCA NN approach. We show that value-at-risk estimation errors are reduced when the proposed anomaly detection model is used with a basic imputation approach to correct the anomaly.

A Full-Dimensional ab initio Intermolecular Potential Energy Surface and Dipole Moment Surfaces for H2O-Ar

Background: The H2O–Ar system has attracted significant interest in recent years because it is an important model to study inelastic scattering between atoms and triatomic molecules. A high-accuracy intermolecular potential energy surface (IPES) is the foundation for theoretical study on molecular collision dynamics for H2O–Ar. In addition, dipole moment surfaces (DMSs) are one of the prerequisites for spectral simulation. Objective: This study aimed to obtain a full-dimensional intermolecular potential energy surface and dipole moment surfaces for the van der Waals complex H2O–Ar. Methods: In this study, ab initio energy points were computed at the frozen-core (FC) explicitly correlated coupled-cluster [FC-CCSD(T)-F12a] level, with the augmented correlation-consistent polarized valence quadruple-zeta basis set plus bond functions. The permutation invariant polynomial neural network (PIP-NN) approach is adopted to fit the IPES, while the DMSs are constructed at the MP2/AVTZ level and fitted by the NN approach. Results: With a root-mean-square-error (RMSE) of 0.284 cm-1, the IPES can accurately describe the motion of the H2O–Ar complex between R = 4 and 20 a0 in the energy range up to 10000 cm-1. The fitting errors of all the data points are 6.192 and 6.509 mDebye for the X and Z components, respectively. The global minimum of -140.633 cm-1 has the plane geometry, while the dipole moment of H2O–Ar is 1.853 Debye at the equilibrium structure. Conclusion: In summary, we report a full-dimensional intermolecular potential energy surface for H2O–Ar. The IPES precisely reproduces CCSD(T)-F12a electronic energies with a large basis set. The corresponding dipole moment surfaces have also been reported. In comparison with previous work, the employment of the high-level ab initio method will make our IPES more reliable. Several typical 2D contour plots of the IPES and DMSs are also shown. The argon atom has a weak effect on the dipole moment of the H2O–Ar complex. The FORTRAN codes to generate 6D potentials and dipole moments reported here are available on request from the authors.

Feed-forward neural networks for fitting of kinetic energy and its functional derivative

We present a feed-forward NN approach for fitting of kinetic energy and its functional derivative for a 1-dimensional system. The density is represented in terms of orthogonal basis functions. The coefficients of basis functions forms the inputs to NN. Using this approach we found no oscillatory behaviour in functional derivative. Using NN based functional derivative we determine the ground-state density by solving the Euler–Lagrange equation. The presented approach can open up new ways for accurate calculations kinetic energy and the functional derivative, which can be considered as an important step towards advancement of machine learning based OF-DFT methods.

Solving the initial value problem of ordinary differential equations by Lie group based neural network method.

To combine a feedforward neural network (FNN) and Lie group (symmetry) theory of differential equations (DEs), an alternative artificial NN approach is proposed to solve the initial value problems (IVPs) of ordinary DEs (ODEs). Introducing the Lie group expressions of the solution, the trial solution of ODEs is split into two parts. The first part is a solution of other ODEs with initial values of original IVP. This is easily solved using the Lie group and known symbolic or numerical methods without any network parameters (weights and biases). The second part consists of an FNN with adjustable parameters. This is trained using the error back propagation method by minimizing an error (loss) function and updating the parameters. The method significantly reduces the number of the trainable parameters and can more quickly and accurately learn the real solution, compared to the existing similar methods. The numerical method is applied to several cases, including physical oscillation problems. The results have been graphically represented, and some conclusions have been made.

<b>Fast Calculation Method of Abnormality Degree for Real Time Abnormality Detection in Vehicle Equipment</b>

Vibration monitoring is effective for early detection of equipment failure. In the vibration monitoring system proposed in this paper, abnormality detection is performed by applying the nearest neighbor method (NN) to the octave band analysis results of vibration. However, the NN requires a long calculation time and is not suitable for detecting abnormalities in real time. Therefore, applying the One Class Support Vector Machine (OCSVM) to abnormality detection was considered. In this paper, the OCSVM was applied to actual vibration data, and the calculation time was compared with those of the NN. The result shows that the calculation time is significantly reduced compared to the NN approach.

K-NEAREST NEIGHBOR AND LINEAR REGRESSION IN THE PREDICTION OF THE ARTIFICIAL FORM FACTOR

k-nearest neighbor and linear regression in the prediction of the artificial form factor . T he proposal of this study was to test whether the performance of the non parametric approach k -Nearest Neighbor ( k -NN), would improve estimates of individual artificial form factor ( f 1.3 ) of trees of the hybrid Eucalyptus urophylla x Eucalyptus grandis compared to the O rdinary L east S quares method. A total of 149 sample-trees were selected , felled , and diameter was measured along the trunk at 10% (d 0. 1 ), 30% (d 0. 3 ), 50% (d 0. 5 ) and 70% (d 0. 7 ) of commercial height and posteriorly at 2m intervals . Mathematical models recognized in the literature for predicting the form factor were adjusted for comparison . The hyperparameter k of optimum adjust ment for the k -NN estimator was obtained by repeated cross-validation. The training data of the k -NN regression model were identical to those used in the adjustment of the linear regression models since m ost multiple linear regression models present problems of collinearity or multicollinearity. The use of the covaria te (d0.3.d0.7)/d1.32 and k = 15 made it possible to construct k -NN models with better generalization capacity. The potential of the k -NN estimator to predict the artificial form factor and thus to obtain less biased estimates of individual tree volumes was demonstrated and considered to be superior to the use of linear regression and average form factors. The k -NN approach can be considered more generic for predicti on of the tree form factor , and its use is recommended when classical linear regression models or other simpler methods do not yield good results.

Neural Network Technique for Diabetic Retinopathy Detection

The diabetes retinopathy is the application of medical image processing. The retinal images are evaluated to diagnose the DR. It is however, time consuming and resource demanding to manually grade the images such that the severity of DR can be defined. When the tiny blood vessels present within the retina are damaged, only then can one notice this problem. Blood will flow from this tiny blood vessel and features are formed from the fluid that exists on retina. The kinds of features involved here due to the leakage of fluid and blood from the blood vessels are considered to be the most important factors to study this problem. The diabetes retinopathy detection techniques has the three phase which pre-processing, segmentation and classification. In this work, NN approach is used for the classification of diabetes portion from the image. The proposed model is implemented in MATLAB and results are analyzed in terms of certain parameters.

Novel relative relevance score for estimating brain connectivity from fMRI data using an explainable neural network approach

BackgroundFunctional integration or connectivity in brain is directional, non-linear as well as variable in time-lagged dependence. Deep neural networks (DNN) have become an indispensable tool everywhere, by learning higher levels of abstract and complex patterns from raw data. However, in neuroscientific community they generally work as black-boxes, leading to the explanation of results difficult and less intuitive. We aim to propose a brain-connectivity measure based on an explainable NN (xNN) approach. New MethodWe build a NN-based predictor for regression problem. Since we aim to determine the contribution/relevance of past data-point from one region i in the prediction of current data-point from another region j, i.e. the higher-order connectivity between two brain-regions, we employ layer-wise relevance propagation (Bach et al., 2015) (LRP, a method for explaining DNN predictions), which has not been done before to the best of our knowledge. Specifically, we propose a novel score depending on weights as a quantitative measure of connectivity, called as relative relevance score (xNN-RRS). The RRS is an intuitive and transparent score. We provide an interpretation of the trained NN weights with-respect-to the brain-connectivity. ResultsFace validity of our approach is demonstrated with experiments on simulated data, over existing methods. We also demonstrate construct validity of xNN-RRS in a resting-state fMRI experiment. ComparisonOur approach shows superior performance, in terms of accuracy and computational complexity, over existing state-of-the-art methods for brain-connectivity estimation. ConclusionThe proposed method is promising to serve as a first post-hoc explainable NN-approach for brain-connectivity analysis in clinical applications.

The Passive Microwave Neural Network Precipitation Retrieval (PNPR) Algorithm for the CONICAL Scanning Global Microwave Imager (GMI) Radiometer

This paper describes a new rainfall rate retrieval algorithm, developed within the EUMETSAT H SAF program, based on the Passive microwave Neural network Precipitation Retrieval approach (PNPR v3), designed to work with the conically scanning Global Precipitation Measurement (GPM) Microwave Imager (GMI). A new rain/no-rain classification scheme, also based on the NN approach, which provides different rainfall masks for different minimum thresholds and degree of reliability, is also described. The algorithm is trained on an extremely large observational database, built from GPM global observations between 2014 and 2016, where the NASA 2B-CMB (V04) rainfall rate product is used as reference. In order to assess the performance of PNPR v3 over the globe, an independent part of the observational database is used in a verification study. The good results found over all surface types (CC > 0.90, ME < −0.22 mm h−1, RMSE < 2.75 mm h−1 and FSE% < 100% for rainfall rates lower than 1 mm h−1 and around 30–50% for moderate to high rainfall rates), demonstrate the good outcome of the input selection procedure, as well as of the training and design phase of the neural network. For further verification, two case studies over Italy are also analysed and a good consistency of PNPR v3 retrievals with simultaneous ground radar observations and with the GMI GPROF V05 estimates is found. PNPR v3 is a global rainfall retrieval algorithm, able to optimally exploit the GMI multi-channel response to different surface types and precipitation structures, that provide global rainfall retrieval in a computationally very efficient way, making the product suitable for near-real time operational applications.

A <italic>k</italic>-NN-Based Localization Approach for Crowdsourced Air Traffic Communication Networks

In this paper, we argue that current state-of-the-art methods of aircraft localization such as multilateration are insufficient, in particular for modern crowdsourced air traffic networks with random, unplanned deployment geometry. We propose an alternative, a grid-based localization approach using the k -nearest neighbor ( k -NN) algorithm, to deal with the identified shortcomings. Our proposal does not require any changes to the existing air traffic protocols and transmitters, and is easily implemented using only low-cost, commercial-off-the-shelf hardware. Using an algebraic multilateration algorithm for comparison, we evaluate our approach using real-world flight data collected with our collaborative sensor network OpenSky. We quantify its effectiveness in terms of aircraft location accuracy, surveillance coverage, and the verification of false position data. Our results show that the grid-based k -NN approach can increase the effective air traffic surveillance coverage compared to multilateration by a factor of up to 2.5. As it does not suffer from dilution of precision to the same extent, it is more robust in noisy environments and performs better in pre-existing, unplanned receiver deployments. We further find that the mean aircraft location accuracy can be increased by up to 41% in comparison with multilateration while also being able to pinpoint the origin of potential spoofing attacks conducted from the ground.

A combined NN and dynamic gain-based approach to further stabilize nonlinear time-delay systems

This paper focuses on the state-feedback control problem for a class of high-order nonlinear systems with unknown time delay and control coefficients. Based on a novel dynamic gain-based backstepping technique and radial basis function neural network (RBF NN) approximation approach, the restrictions on high-order and nonlinearities are removed or further relaxed. Under these weaker conditions, a smooth state-feedback controller is skillfully constructed with only one adaptive parameter. In addition, the knowledge of time delay, NN nodes and weights is not necessary to be known a priori. It is proven that the designed controller can render the closed-loop system be semi-globally uniformly ultimately bounded. Finally, both practical and numerical examples are shown to demonstrate the effectiveness of the proposed scheme.

Classifier Architectures for Acoustic Scenes and Events: Implications for DNNs, TDNNs, and Perceptual Features from DCASE 2016

This paper evaluates neural network NN based systems and compares them to Gaussian mixture model GMM and hidden Markov model HMM approaches for acoustic scene classification SC and polyphonic acoustic event detection AED that are applied to data of the “Detection and Classification of Acoustic Scenes and Events 2016” DCASE'16 challenge, task 1 and task 3, respectively. For both tasks, the use of deep neural networks DNNs and features based on an amplitude modulation filterbank and a Gabor filterbank GFB are evaluated and compared to standard approaches. For SC, additionally a time-delay NN approach is proposed that enables analysis of long contextual information similar to recurrent NNs but with training efforts comparable to conventional DNNs. The SC system proposed for task 1 of the DCASE'16 challenge attains a recognition accuracy of 77.5%, which is 5.6% higher compared to the DCASE'16 baseline system. For the AED task, DNNs are adopted in tandem and hybrid approaches, i.e., as part of HMM-based systems. These systems are evaluated for the polyphonic data of task 3 from the DCASE'16 challenge. Several strategies to address the issue of polyphony are considered. It is shown that DNN-based systems perform less accurate than the traditional systems for this task. Best results are achieved using GFB features in combination with a multiclass GMM-HMM back end.