OPTIMAL HIDDEN UNITS FOR TWO-LAYER NONLINEAR FEEDFORWARD NEURAL NETWORKS

This chapter describes parallel processing of robot-arm control computation and simulation. The parallel processing of robot control computation has attracted much attention to develop a cost-effective, compact, and advanced robot controller which allows a robot system to perform very complicated operations quickly and accurately even in hazardous environments, such as space and ocean floor. Also, parallel processing of robot motion simulation is important to efficiently develop the advanced robot hardwares and the controllers. This chapter introduces several parallel processing schemes on multiprocessor systems for the robot-arm control computation and the robot-arm simulation. More concretely, first of all, parallel processing schemes for the Newton-Euler equations for dynamic control are introduced. Secondly, how to implement the schemes on an actual multiprocessor system is discussed. Thirdly, parallel processing schemes for Walker & Orin’s algorithm for dynamics simulation and an implementation of the schemes on a multiprocessor system are described. Finally, future directions of the parallel processing of the robot control and simulation are briefly discussed.

В статті розглянуто задачу прогнозування дискретних значень потужності споживання електричної енергії, яка вирішується з використанням одного з методів штучних нейронних мереж - обмеженої машини Больцмана. В цьому методі прогнозування здійснюється шляхом підлаштування вагових функцій шарів нейронної мережі таким чином, щоб мінімізувати відстань між реальними та розрахунковими значеннями. У кожній фазі навчання мережі за допомогою відповідних математичних функцій розраховуються і змінюються коефіцієнти ваг з урахуванням їх поточних значень. В результаті після навчання система здатна ефективно прогнозувати часові залежності потужності споживання електричної енергії, що є важливим при розробці системі керування електроспоживанням промислових та побутових об’єктів.

Stability analysis of a three-term backpropagation algorithm

Considers the problem of modeling and controlling the impact of a two-degree-of-freedom planar robot arm against an infinitely rigid and massive surface. For this case study, the basic equations describing the motion of the robot arm are derived for noncontact and contact conditions. A control scheme is proposed on the basis of a reduced-order observer that is able to asymptotically estimate the impact-induced forces and to allow their asymptotic compensation when the robot arm is in contact with the surface. The resulting control system ensures the global asymptotic regulation of the position of the arm with an assigned impact-induced force. The case study is completed by a simulation test.

Generative models offer advantageous characteristics for classification tasks, such as the availability of unsupervised data and calibrated confidence. In contrast, discriminative models have advantages in terms of their potential to outperform their generative counterparts and the simplicity of their model structures and learning algorithms. In this article, we propose a method to train a hybrid of discriminative and generative models in a single neural network (NN), which exhibits the characteristics of both models. The key idea is the Gaussian-coupled softmax layer, which is a fully connected layer with a softmax activation function coupled with Gaussian distributions. This layer can be embedded into an NN-based classifier and allows the classifier to estimate both the class posterior distribution and the input data distribution. We demonstrate that the proposed hybrid model can be applied to semi-supervised learning and confidence calibration.

In view of present advancements in computing, with the development of distributed environment, many problems have to deal with distributed input data where individual data privacy is the most important issue to be addressed, for the concern of data owner by extending the privacy preserving notion to the original learning algorithms. Privacy Preserving Data Mining has become an active research area in addressing various privacy issues while bringing out solutions for them. There has been lot of progress in developing secure algorithms and models, able to preserve privacy using various data mining techniques like association, classification and clustering, where as importance of privacy preserving techniques applied for learning algorithms related to neural networks for mining problems are still in infancy. Our work in this paper focused on preserving privacy of an individual, using self organizing map (SOM) adopted for collaborative clustering of distributed data between multiple parties. We present Privacy Preserving Collaborative Clustering method using SOM (PPCSOM) for Horizontal Data Distribution, which allows multiple parties perform clustering in a collaborative approach using SOM neural network, without revealing their data directly to each other, in order to preserve privacy of all parties. Our simulation results shows that implementation of PPCSOM method achieves comparable accuracy and assured Privacy than the original non privacy preserving SOM algorithm.

The present study is a real-time application of Artificial Neural Networks (ANNs) to estimate, predict and forecast the suspended sediment transport in streams and river systems using time series data. NN approach is a crucial and readily adaptable important methodology when hydrograph/unit hydrograph methodology and conventional mathematics are inconvenient to apply for an emergency situation to act as the sediment runoff time series data is highly complex in nature i.e. situations like availability of less amount of data, sometimes the very high and huge quantity of data, poor quality of data, erratic nature of data, need of emergency prediction and forecasting, etc. The present study is pertaining to a tributary; namely Peddavagu in the Godavari river system in India. A back propagation algorithm was employed in the methodology among the layers of multi-layered feed-forward neural network with one hidden layer and two hidden layer approaches. The study also demonstrates the beauty of an application of computational intelligence/ artificial intelligence technique as how it is superior to the conventional mathematical modeling. Sediment rating curve (SRC), mathematical modeling was carried out to present a comparative performance of the applied methodology. Statistical performance evaluation criteria using root mean square error (RMSE), correlation coefficient (CC) and coefficient of determination (DC) comprehend the sensitivity in applying the ANNs for complex suspended sediment time series to predict and forecast the sediment hydrology of river systems. Hysteresis effect of sediment runoff is also a component in the study to reveal a result of many number of sediment concentration magnitudes for a specific runoff magnitude in the river flow system.

Symmetry is present in nature and science. In image processing, kernels for spatial filtering possess some symmetry (e.g. Sobel operators, Gaussian, Laplacian). Convolutional layers in artificial feed-forward neural networks have typically considered the kernel weights without any constraint. We propose to investigate the impact of a symmetry constraint in convolutional layers for image classification tasks, taking our inspiration from the processes involved in the primary visual cortex and common image processing techniques. The goal is to determine if it is necessary to learn each weight of the filters independently, and the extent to which it is possible to enforce symmetrical constraints on the filters throughout the training process of a convolutional neural network by modifying the weight update preformed during the backpropagation algorithm and to evaluate the change in performance. The symmetrical constraint reduces the number of free parameters in the network, and it is able to achieve near identical performance. We address the following cases: x/y-axis symmetry, point reflection, and anti-point reflection. The performance is evaluated on four databases of images representing handwritten digits. The results support the conclusion that while random weights offer more freedom to the model, the symmetry constraint provides a similar level of performance while decreasing substantially the number of free parameters in the model. Such an approach can be valuable in phase-sensitive applications that require a linear phase property throughout the feature extraction process.

Ant colony optimization (ACO) algorithm has the powerful ability of searching the global optimal solution, and backpropagation (BP) algorithm has the feature of rapid convergence on the local optima. The proper hybrid of the two algorithms (ACO-BP) may accelerate the evolving speed of neural networks and improve the forecasting precision of the well-trained networks. ACO-BP scheme adopts ACO to search the optimal combination of weights in the solution space, and then uses BP algorithm to obtain the accurate optimal solution quickly. The ACO-BP and BP algorithms were applied to the problems of function approaching and modeling quantitative structure-activity relationships of Herbicides. Experiment results show that the proposed ACO-BP scheme is more efficient and effective than BP algorithm. Furthermore, ACO-BP reliably performs well when the number of hidden nodes varies.KeywordsHide NodeEvolve Neural NetworkForecast PrecisionPheromone TablePheromone IntensityThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

A Statistical Approach to Learning and Generalization in Layered Neural Networks

A variable structure model following control (VSMFC) design method is proposed for robotics applications. The method allows the designer to model the nonlinear dynamic interactions among the different mechanical components of the robot system even when an adaptive control solution is called for to satisfy stringent performance requirements. A VSMFC design for a planar two-link robot arm is considered as an example. >

Origin Destination (OD) Traffic Matrix (TM) contains the number of packets transferring among the nodes at a particular time in a confined network. This TM can be used to schedule traffic, manage peak traffic, make network rules, identify suspicious activities, traffic routing, etc., for which the prediction should be more accurate. Traditionally Machine Learning (ML) algorithms have played a crucial role in predicting the time series data such as TMs. In this paper, four different ML algorithms, namely Long Short-Term Memory (LSTM), Gate Recurrent Unit (GRU), Simple Recurrent Neural Network (SRNN), and Bidirectional LSTM (BiLSTM), are applied, which use recurrent layers in neural networks, to the three real-world TM data sets such as American Research and Education Network (Abilene), Europe Research and Education Network (GÉANT), China Education and Research Network (CERNET) and a custom testbed generated TM data set. The final results show that the tested algorithms make predictions with minimum errors.

This paper is proposed to solve the inverse kinematic (IK) problem of two-degree-of-freedom planar robot arm using neural networks (NN). Several NN model designs of distinct hidden neurons based on the sum of square error function of joint angle are developed and trained with generalised reduced gradient algorithm. The paper is also intended to demonstrate the modelling process of feed-forward NN topology in spreadsheet environment. The spreadsheet functions as INDEX, SUMPRODUCT, EXP, and SUMSQ; the utilities as name manager, data validation, data table, ActiveX controls, answer report, and charts; and the add-in Solver are utilised to develop the models. With the input parameters of link lengths and end-effector position and orientation, two models with the structures 5-12-1 and 5-10-1 are discovered best-capable in predicting first and second joint angles respectively. This NN-based IK technique contributes significantly to the optimal motion control of robot arm for quality processing and assembly tasks.

Using results on the topology of moduli space of polygons (Jaggi, Ph.D. Thesis, University of Bern, 1992; Kapovich and Millson, J. Differ. Geom. 42:133---164, 1994), it can be shown that for a planar robot arm with n segments there are some values of the base-length, z (i.e., length of line joining the base of the arm with its end-effector), at which the configuration space of the constrained arm (arm with its end effector fixed) has two disconnected components, while at other values it has one connected component. We first review some of these known results relating the value of z with the connectivity of the constrained configuration space. Then the main design problem addressed in this paper is the construction of pairs of continuous inverse kinematics for arbitrary robot arms, with the property that the two inverse kinematics agree (i.e. return the same configuration) when the constrained configuration space has a single connected component, but they give distinct configurations (one in each connected component) when the configuration space of the constrained arm has two components. This design is made possible by a fundamental theoretical contribution in this paper--a classification of configuration spaces of robot arms such that the type of path that the system (robot arm) takes through certain critical values of the forward kinematics function is completely determined by the class to which the configuration space of the arm belongs. This classification result makes the aforesaid design problem tractable, making it sufficient to design a pair of inverse kinematics for each class of configuration spaces (three of them in total). The motivation for this work comes from a more extensive problem of motion planning for the end effector of a robot arm, in which the ability to continuously sample one configuration from each connected component of the constrained configuration spaces of the arm enables us to dramatically reduce the dimensionality of the space in which the planning has to be performed, without sacrificing algorithmic completeness. We start the paper with the general motivation, but address only the problem of sampling such configurations when there is no obstacle in the environment--a problem that in itself is non-trivial. We demonstrate the simplicity and the low complexity of the presented algorithm through a Javascript + HTML5 based implementation available at http://hans.math.upenn.edu/~subhrabh/nowiki/robot_arm_JS-HTML5/arm.html.

The authors propose an alternative coordinated motion control architecture for robot arms manipulating an object. The motion and the internal force of the object are resolved into the motion of each arm. Each arm is controlled, on the basis of the virtual internal model, so that they operate in coordination even if geometric errors exist in the arms and the object. The virtual internal model is a reference model driven by sensory information implemented in the controller. The proposed architecture maintains the stability of the system even if breakage of the manipulated object occurs. The control algorithm was experimentally applied to the coordinated motion control of two planar robot arms, each of which has three degrees of freedom. The results illustrate the validity of the proposed control architecture. >