Multivariate Artificial Neural Networks Research Articles

Quantum Dot Light‐Emitting Synaptic Transistor for Parallel Data Transmission of Diverse Artificial Neural Network

AbstractArtificial synaptic devices serve as the cornerstone of artificial neural networks, much research is devoted to the development of artificial synaptic devices with multiple functions for the future construction of large‐scale artificial neural networks. By adding optical signal output to traditional synaptic devices, the strategy of transforming the devices from a single electrical interconnection to an optoelectronic interconnection is considered to be an effective way to solve the problem of wire cross‐talk in large‐scale artificial neural networks. Herein, a quantum‐dot light‐emitting synaptic transistor capable of dual output of optoelectronic signals by integrating the functions of light‐emitting transistor and synaptic transistor into a single device is demonstrated for the first time. Based on the novel working mechanism and the excellent optoelectronic properties of quantum dots, the device can exhibit dual responses of electrical and optical signals under electrical pulse stimulation. More importantly, some key synaptic functions such as excitatory postsynaptic current, paired pulse facilitation, high‐pass filtering properties, and the transition from short‐term memory to long‐term memory are successfully simulated in the device. In addition, classical conditioned reflex experiments as well as the processes of learning and forgetting are optically and electrically simulated. This work provides a feasible way to realize multivariate artificial neural networks with high integration and optoelectronic interconnection to transmit information, showing great potential in the development of neuromorphic computing in the future.

PSVI-16 Comparison of univariate and multivariate machine learning applications for genomic predictions of growth and carcass traits from Brangus heifers

Abstract Data for growth and carcass traits were obtained from 738 Brangus heifers (3/8 Brahman-Bos indicus × 5/8 Angus-Bos taurus) registered with International Brangus Breeders Association. Phenotypes included body weights (BW, WW, and YW) collected at birth, ~205 and 365 d of age for growth traits and yearling ultrasound assessment of longissimus muscle area (LMA), percent intramuscular fat (IMF), and depth of rib fat (FAT) for carcass traits and were used to compare univariate and multivariate artificial neural networks (ANN) models with the learning algorithms (Levenberg–Marquardt (LM), Bayesian Regularization (BR) and Scaled Conjugate Gradient (SCG)) and transfer functions (tangent sigmoid and linear) using 1 to 20 neurons models based on the input from G genomic relationship matrix. Pearson correlation coefficients for evaluating model performances in testing datasets indicated that univariate ANN models resulted in better genomic prediction than the multivariate ANN models regardless of the learning algorithms and transfer functions for carcass traits. However, there was no clear difference between univariate and multivariate ANN models about genomic prediction for growth traits. In multivariate ANN models, the prediction performance of the combination of learning algorithms and transfer functions changed for growth and carcass traits and there were no superior multivariate ANN models with BR, LM and SCG learning algorithms and transfer functions. However, the correlation coefficients from univariate ANN-BR model indicated better genomic prediction than univariate ANN-LM and ANN-SCG models. The application of different transfer functions did not make any significant difference on the genomic prediction performance of ANN models with different learning algorithms. Results of this study suggest the use of univariate ANN models with BR learning algorithm for genomic prediction of growth and carcass traits.

Inversely Calibrated Curvilinear Artificial Neural Network Model for Simultaneous Assay of Ternary Cardiovascular Drug Mixture

Novel chemometric design, tailored for pre-clinical multiple drug screening, goals for bioanalytical future scope. A highly sensitive, non-linear multivariate Artificial Neural Network (ANN) is developed and applied for simultaneous spectrophotometric determination of three commonly concomitant cardiovascular drugs in laboratory made mixtures and spiked human plasma samples. Ticagrelor, Irbesartan and Hydrochlorothiazide have been simultaneously quantified in the curvilinear ranges of 0-30 µg/mL, 0-10 µg/mL and 0-3 µg/mL respectively. Highly overlapping Near UV absorption spectra of three drugs, in the region of 215-280 nm, have been recorded 1-nm range in synthetic ternary mixtures and trained iteratively. By inversely relating the concentration matrix (x-block) with its corresponding absorption one (y-block), gradient-descent back-propagation ANN calibration could be computed and optimized. All proposed mathematical modeling was manipulated using MATLAB® 2007, reaching down to sixth order exponential Mean Square Error, MSE. To validate, independent set of ternary synthetic mixtures has been constructed and examined, where excellent recovery results has been obtained. Furthermore, application of suggested model to varying ratios synthetic ternary mixtures as well as spiked plasma samples has resulted in accurate, precise and robust estimations with no background interference. ANN method was compared to a reference HPLC method; Student’s t-test and F-variance ratio were calculated and showed insignificant difference. This chemometric approach is an eco-friendly green assay, time-saving, and economic method. It initiates a pathway for clinical drug screening through affordable spectroscopic instrumentation.

Temporal transferability of marine distribution models in a multispecies context

Abstract Transferability of species distribution models is crucial for supporting biological conservation and management planning in the face of climate changes; however, there is a lack of understanding in how to achieve transference spatially and temporally, especially in a multispecies circumstance. This study evaluated the roles of model complexity and time scope in temporal transferability for multispecies distribution modelling. We collected data of multispecies distribution from a seven-year marine survey in coastal Yellow Sea, China, and used the data to evaluate the predictive performances of an array of multi-species distribution models (constrained linear ordination, constrained additive ordination, hierarchical models of species communities, multivariate random forests, multivariate tree boosting model, and multivariate artificial neural network), characterized by different structures, levels of complexity and time spans with 36 models and 50 scenarios. Our study showed that model transference was proper for some species but limited at the community level in general, and no specific modelling algorithm (regression models/machine-learning) or level of model complexity was found to be superior. Instead, model structure, features of target species, and time scope contributed substantially and influenced transferability in a contingent way. Temporal transferability tended to increase with the length of training periods and decay with the time gaps of prediction, whereas the magnitude varied among modelling algorithms. In addition, prolonging the time series of survey programs alone may have limited effectiveness on community-level transferability. This study contributes to the understanding of temporal transferability of multi-species distribution models and inform their applications in biological conservation and management from a temporal perspective.

Environmental-economic optimization for implementation of parabolic collectors in the industrial process heat generation: Case study of Mexico

Abstract This paper proposes an economic and environmental analysis for the design of solar heat systems and their incorporation in the industrial sector. A novel computational methodology is implemented to help the sustainable integration of parabolic trough collectors photothermal technology into the low and medium enthalpy industrial activities. A multivariate artificial neural network that involves environmental, operational, and economic aspects was used to transfer the phenomenon under study into a simple and fast computing multi-output mathematical model. The optimal trade-off between environmental benefits and investment viability of the hybrid solar plant’s design is obtained by a multi-objective optimization process. The objective functions consider the maximization in CO2 mitigation and net present value, and the minimization of the total life-cycle cost. The final optimal result is selected by using the TOPSIS decision-making method. Besides, a sensitivity analysis is conducted to report the system performance with respect to back-up boiler fuel type, climate region, the volume of the storage tank, and solar field area. The work contemplates the case study of a solar thermal plant based on parabolic collectors integrated into a pre-existing industrial process in Mexico. The study considered four of the most common climatic regions and the most representative heating fossil fuels in the national industrial sector. Based on the results, the best profitability and CO2 mitigation are achieved in warm climate regions. Moreover, diesel was identified as the most profitable back-up fuel scenario and natural gas as the least viable. Analysis of the energy contribution describes that implementation of parabolic trough collectors in the presented industrial process covers just over 40%–80% of the energy required, depending on the climate region. The presented methodology constitutes a rapid and low-cost computational tool which facilitates decision making for the implementation of solar heat industrial process systems. Moreover, it can be applied to other industrial processes with several photothermal technologies for clean heat generation.

Lane-based short-term urban traffic parameters forecasting using multivariate artificial neural network and locally weighted regression models: A genetic approach

Short-term prediction of traffic conditions on urban arterials has recently become increasingly important because of its vital role in the basic traffic management functions and trip decision-making processes. Such information is useful for optimal infrastructure operation, routing, and trip scheduling. However, forecasting models offering a high accuracy at a fine temporal resolution (e.g., 1 or 5 min) and, especially, lane-based are still rare and need special attention. Given the dynamic and stochastic nature of traffic, this study proposes a genetically optimized artificial neural network (GA-ANN) and locally weighted regression (GA-LWR) multivariate models, for short-term traffic prediction using a combination of multiple traffic variables such as volume, occupancy, and speed, during peak and off-peak periods. The proposed 5-min GA-ANN and GA-LWR disaggregate multivariate models show lower average and 95th percentile (P95) errors, when compared to those reported in the literature. In particular, for peak and off-peak time prediction, the GA-ANN disaggregate multivariate models result in most of the average errors being from 2% to 5% and the 95th percentile errors being from 9% to 10%. On the other hand, for peak and off-peak time traffic prediction, the GA-LWR disaggregate multivariate models show that most of the average errors are lower than 5% and the 95th percentile errors are lower than 10%. Meanwhile, for peak and off-peak time prediction, both GA-ANN and GA-LWR disaggregates models show lower MSE of 0.11–1.84. Hence, such techniques are believed useful for developing a robust urban traffic forecasting system.

Forecasting Inflation in India

This paper presents an application of Artificial Neural Network (ANN) to forecast inflation in India during the period 1994-2009. The study presents four different ANN models on the basis of inflation (WPI), economic growth (IIP), and money supply (MS). The first model is a univariate model based on past WPI only. The other three are multivariate models based on WPI and IIP, WPI and MS, WPI, and IIP and MS. In each case, the forecasting performance is measured by mean squared errors and mean absolute deviations. The paper finally concludes that multivariate models show better forecasting performance over the univariate model. In particular, the multivariate ANN model using WPI, IIP, and MS resulted in better performance than the rest of other models to forecast inflation in India.

A Proposed Algorithm For Multivariate Artificial Neural Network

A Proposed Algorithm For Multivariate Artificial Neural Network

Determining Inputs for Neural Network Models of Multivariate Time Series

In recent years, artificial neural networks have been used successfully to model multivariate water resources time series. By using analytical approaches to determine appropriate model inputs, network size and training time can be reduced. In this paper, it is proposed that the method of Haugh and Box and a new neural network–based approach can be used to identify the inputs for multivariate artificial neural network models. Both methods were used to obtain the inputs for a multivariate artificial neural network model used for forecasting salinity in the River Murray at Murray Bridge, South Australia. The methods were compared with a third method that uses knowledge of travel times in the river to identify a reasonable set of inputs. The results obtained indicate that all three methods are suitable for determining the inputs for multivariate time series models. However, the neural network–based method is preferable because it is quicker and simpler to use. Any prior knowledge of the underlying processes should be used in conjunction with the neural network method.

Neural Network Process Models Based on Linear Model Structures

The KBANN (Knowledge-Based Artificial Neural Networks) approach uses neural networks to refine knowledge that can be written in the form of simple propositional rules. This idea is extended by presenting the MANNIDENT (Multivariable Artificial Neural Network Identification) algorithm by which the mathematical equations of linear dynamic process models determine the topology and initial weights of a network, which is further trained using backpropagation. This method is applied to the task of modeling a nonisothermal chemical reactor in which a first-order exothermic reaction is occurring. This method produces statistically significant gains in accuracy over both a standard neural network approach and a linear model. Furthermore, using the approximate linear model to initialize the weights of the network produces statistically less variation in model fidelity. By structuring the neural network according to the approximate linear model, the model can be readily interpreted.

Using Knowledge-Based Neural Network Process Models for Model-Based Control

This paper describes the MANNCON (Multivariable Artificial Neural Network Control) system in which knowledge-based artificial neural network models are incorporated into model-based controllers. Designs using Internal Model Control (IMC) and Direct Synthesis Control (DSC) as well as a supervisory ANN-based controller are presented. These methods are applied to the task of controlling a nonisothermal CSTR in which a first-order exothermic reaction is occurring. The nonlinear ANN-based controllers perform much better than their linear counterparts in controlling the process over a wide range of conditions.

Experiences with model-based controllers based on neural network process models

An earlier paper introduced the MANNIDENT (multivariable artificial neural network identification) approach to process modelling in which approximate linear models determine the topology and initial weights of a neural network and which provides a structure for interpretation of the final model. This paper introduces the MANNCON (multivariable artificial neural network control) system, in which these models are incorporated into model-based controllers. Designs using internal model control (IMC) and direct synthesis control (DSC) as well as a supervisory ANN-based controller are presented. These methods are applied to the task of controlling a non-isothermal CSTR in which a first-order exothermic reaction is occurring. The nonlinear ANN-based controllers perform much better than their linear counterparts in controlling the process over a wide range of conditions. Furthermore, the superivosyr ANN controller showed better robustness properties with respect to disturbance rejection and to plant-model mismatch than the other controllers studied.

Creating efficient nonlinear neural network process models that allow model interpretation

Abstract The KBANN (knowledge-based artificial neural networks) approach uses neural networks to refine knowledge that can be written in the form of simple propositional rules. This idea is extended by presenting the MANNIDENT (multivariable artificial neural network identification) algorithm by which the mathematical equations of linear process models determine the topology and initial weights of a network, which is further trained using backpropagation. This method is applied to the task of modelling a non-isothermal CSTR in which a first-order exothermic reaction is occurring. This method produces statistically significant gains in accuracy over both a standard neural network approach and a linear model. Furthermore, using the approximate linear model to initialize the weights of the network produces statistically less variation in accuracy. By structuring the neural network according to the approximate linear model, the model can be readily interpreted.

Refining PID Controllers Using Neural Networks

The KBANN (Knowledge-Based Artificial Neural Networks) approach uses neural networks to refine knowledge that can be written in the form of simple propositional rules. We extend this idea further by presenting the MANNCON (Multivariable Artificial Neural Network Control) algorithm by which the mathematical equations governing a PID (Proportional-Integral-Derivative) controller determine the topology and initial weights of a network, which is further trained using backpropagation. We apply this method to the task of controlling the outflow and temperature of a water tank, producing statistically significant gains in accuracy over both a standard neural network approach and a nonlearning PID controller. Furthermore, using the PID knowledge to initialize the weights of the network produces statistically less variation in test set accuracy when compared to networks initialized with small random numbers.