Search

This research is concerned with finding the optimal solution of a job shop scheduling problem (JSSP) in as short time as possible. A JSSP can be formulated as a 0-1 mixed integer programming (0-1 MIP), and it is expected that the optimal schedule can be obtained in a shorter time by predicting a good solution based on data of scheduling performed previously and using the solution as the initial solution for the optimization algorithm. In this concept, it is an important point to predict a solution which satisfy constraints of the 0-1 MIP. This paper provides an improved method based on this concept where the prediction is carried out so that the constraints are always satisfied. Numerical experiments showed that solution time of the proposed method is shorter than that of the previous method, which does not assure the constraints are satisfied, and learning time reduces about half. In addition, it turned out that there is a reason of the shorter solution time other than that the predicted solution always satisfies the constraints. One possible reason is that the number of circular sequences tends to be smaller than the previous method. Further analysis and additional evaluations from this point of view will be performed in a future work. [ABSTRACT FROM AUTHOR]

The paper observes a similarity between the stochastic optimal control of discrete dynamical systems and the learning multilayer neural networks. It focuses on contemporary deep networks with nonconvex nonsmooth loss and activation functions. The machine learning problems are treated as nonconvex nonsmooth stochastic optimization problems. As a model of nonsmooth nonconvex dependences, the so-called generalized-differentiable functions are used. The backpropagation method for calculating stochastic generalized gradients of the learning quality functional for such systems is substantiated basing on Hamilton–Pontryagin formalism. Stochastic generalized gradient learning algorithms are extended for training nonconvex nonsmooth neural networks. The performance of a stochastic generalized gradient algorithm is illustrated by the linear multiclass classification problem. [ABSTRACT FROM AUTHOR]

Abstract Machine learning (ML) techniques play a crucial role in today's modern world. Over the last years, road traffic safety is one of the applications where ML‐methods have been successfully employed to prevent road users from being killed or seriously injured. A reliable data‐driven predictive model is essential for this purpose. This could be achieved by successfully applying an intelligent transportation system to identify a driver at a higher risk of crashes. This study investigates the capabilities of different deep learning techniques to predict motorcycle crash severity. This study is based on 2,430 motorcycle crashes in a mountainous area in the United States over a 10‐year period. Different deep networks (DNNs), including deep belief network, standard recurrent neural network (RNN), multilayer neural network, and single‐layer neural network, were considered and compared in terms of prediction accuracy of motorcycle crash severity. Before conducting any analysis, feature reduction was performed to identify the optimal number of variables to include in the models by minimizing the error rate. Different metrics including the area under the curve and confusion matrix were used to compare the different models. Although the analyses were conducted on a relatively small dataset, the results indicate that almost all the DNN models better perform in predicting the severity of motorcycle crashes, compared with the single layer neural network. Finally, the RNN outperforms the other three neural network models. A comprehensive discussion has been made about the methodological approach implemented in this study.

It is assumed that by 2021, about half of the companies will use digital counterparts of different levels. The simplest digital twin models may not use machine learning, but the models using machine learning algorithms will have the greatest advantage. In this article, we offer our approach to the construction of digital twins for real objects. We rely on our unified process of constructing approximate solutions of boundary value problems for equations of mathematical physics and accumulated experience in solving numerous specific problems of this type. In this paper, we present five approaches to the construction of digital twin models based on the evolutionary algorithms developed and tested by us. The peculiarity of our approach to evolutionary algorithms is the use of genetic procedures for constructing the structure of the model and nonlinear optimization algorithms for adjusting its parameters. Also, we propose our approach to the construction of multilayer models upon differential equations, which allows doing without the time-consuming procedure of neural networks training. We are confident that the proposed approaches can significantly simplify and unify the creation and adaptation (keeping up to date) of digital twins for real objects of various kinds – technical, biological, socio-economic, etc.

Machine learning (ML) techniques play a crucial role in today's modern world. Over the last years, road traffic safety is one of the applications where ML‐methods have been successfully employed to prevent road users from being killed or seriously injured. A reliable data‐driven predictive model is essential for this purpose. This could be achieved by successfully applying an intelligent transportation system to identify a driver at a higher risk of crashes. This study investigates the capabilities of different deep learning techniques to predict motorcycle crash severity. This study is based on 2,430 motorcycle crashes in a mountainous area in the United States over a 10‐year period. Different deep networks (DNNs), including deep belief network, standard recurrent neural network (RNN), multilayer neural network, and single‐layer neural network, were considered and compared in terms of prediction accuracy of motorcycle crash severity. Before conducting any analysis, feature reduction was performed to identify the optimal number of variables to include in the models by minimizing the error rate. Different metrics including the area under the curve and confusion matrix were used to compare the different models. Although the analyses were conducted on a relatively small dataset, the results indicate that almost all the DNN models better perform in predicting the severity of motorcycle crashes, compared with the single layer neural network. Finally, the RNN outperforms the other three neural network models. A comprehensive discussion has been made about the methodological approach implemented in this study. [ABSTRACT FROM AUTHOR]

Problems of nonsmooth nonconvex dynamic optimization, optimal control (in discrete time), including feedback control, and machine learning are considered from a common point of view. An analogy between controlling discrete dynamical systems and multilayer neural network learning problems with nonsmooth objective functionals and connections is traced. Methods for computing generalized gradients for such systems based on the Hamilton–Pontryagin functions are developed. Gradient (stochastic) algorithms for optimal control and learning are extended to nonconvex nonsmooth dynamic systems. [ABSTRACT FROM AUTHOR]

Copyright of Cybernetics & Systems Analysis / Kibernetiki i Sistemnyj Analiz is the property of V.M. Glushkov Institute of Cybernetics of NAS of Ukraine and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Tailoring the muckpile shape and its fragmentation to the requirements of the excavating equipment in surface mines can significantly improve the efficiency and savings through increased production, machine life and reduced maintenance. Considering the various blast parameters together to predict the throw is subtle and can lead to wrong conclusions. In this paper, a different approach was followed to combine the representational power of multilayer neural networks and various machine learning techniques to predict the throw of a bench blast using the data from a limestone mine located in central India. Then, using various analysis techniques, the training parameters have been adjusted to reduce the cross-validation error and increase the accuracy. Here, four different architectures of neural networks have been trained by different techniques, and the best model has been selected. The different machine learning techniques have been implemented on the basis of accuracy of the output. The sensitivity analysis has been done to get the relative importance of the variables in prediction of the output. [ABSTRACT FROM AUTHOR]

In this paper, an innovative multilayer neuroadaptive controller is firstly proposed to further improve the output force tracking performance of electro-hydraulic load simulators with uncertainty rejection. Significantly, a highly nonlinear load simulator driven by the single-rod electro-hydraulic actuators is considered, which has universal research significance in relation to the relevant studies. Moreover, smooth endogenous uncertainties, non-smooth endogenous uncertainties in the load pressure dynamics, and exogenous disturbances can all be effectively compensated in a feedforward way, which makes it possess higher load simulation accuracy and wider applicability to working conditions. In detail, multilayer neural networks and extended state observer are employed to estimate endogenous uncertainties and exogenous disturbances, respectively. Specially, the neural network weights are on-line updated via the composite errors comprised of the force tracking and estimate errors, which can increase the resulting learning ability. The closed-loop stability is guaranteed via the strict Lyapunov stability analysis and extensive comparative application results are achieved to validate the high-performance efficacy of the proposed intelligent learning algorithm. • The highly nonlinear load simulator driven by single-rod hydraulic actuators is considered. • Smooth and non-smooth endogenous uncertainties, and exogenous disturbances can be addressed. • The learning ability of neural networks can be increased. [ABSTRACT FROM AUTHOR]

Abstract: This work explores annealed cooperative–competitive learning of multiple modules of Mahalanobis normalized radial basis functions (NRBF) with applications to nonlinear function approximation and chaotic differential function approximation. A multilayer neural network is extended to be composed of multiple Mahalanobis-NRBF modules. Each module activates normalized outputs of radial basis functions, determining Mahalanobis radial distances based on its own adaptable weight matrix. An essential cooperative scheme well decomposes learning a multi-module network to sub-tasks of learning individual modules. Adaptable network interconnections are asynchronously updated module-by-module based on annealed cooperative–competitive learning for function approximation under a physical-like mean-field annealing process. Numerical simulations show outstanding performance of annealed cooperative–competitive learning of a multi-module Mahalanobis-NRBF network for nonlinear function approximation and long term look-ahead prediction of chaotic time series. [Copyright &y& Elsevier]

In this paper, we propose a new network element, “distance-based cosine element,” for neural networks. We also derive a learning algorithm based on the backpropagation algorithms for multilayer networks, The distance-based cosine element inputs a squared distance between an input pattern vector and its weight vector, and uses an affine transformation of cosine function as its output function. The proposed distance-based cosine network is able to improve its learning speed as well as convergence rate because its output function does not have any saturated regions which cause slow learning speed of the backpropagation learning using sigmoid elements. We demonstrate the advantages of the proposed network by solving N-bit parity problems and Fisher's iris classification problem. Experimental results indicate that our distance-based cosine network consistently obtains better results than the conventional sigmoid network in terms of both the learning speed and the convergence rate. © 1999 Scripta Technica, Electr Eng Jpn, 129(4): 87–95, 1999 [ABSTRACT FROM AUTHOR]