The OWE Artificial Neural Network and its Parallel Implementation

نویسندگان

  • Nicolas Pican
  • Yannick Lallement
چکیده

RÉSUMÉ. Les perceptrons multi-couches (ou perceptrons) sont largement utilisés pour l’approximation de fonctions, mais sont très demandeursen temps de calcul. Malheureusement, l’implantation parallèle des perceptrons est un problème complexe ; dans cet article, nous proposons une nouvelle méthode efficace pour leur parallélisation. Nous présentons l’architecture OWE (Orthogonal Weight Estimator), dont les performances sont au moins aussi bonnes que celles d’un perceptron classique avec le même nombre de paramètres libres. L’architecture OWE consiste en un perceptron principal dans lequel chacun des poids est calculé par un perceptron indépendant (un OWE). Le nombre d’OWEs est égal au nombre de poids dans le perceptron principal. Les calculs dans chaque OWE peuvent être menés de façon indépendante, ce qui permet une parallélisation aisée des phases d’apprentissage et de relaxation. Nous décrivons notre implantation de cette architecture sur une machine Intel Paragon, et nous la comparons avec son implantation sur une machine séquentielle. ABSTRACT. Multi-layer perceptrons (MLPs) are a useful and widely used tool for function approximation, but they are very demanding on computer time, hence the need for parallelization. Unfortunately, the parallel implementation of MLPs is not straightforward. In this paper, we propose a new efficient way to parallelize MLPs. We present the OWE (Orthogonal Weight Estimator) architecture, which performs at least as well as a classical MLP with the same number of free parameters. The OWE architecture consists of a main MLP in which the value of each weight is computed by another MLP (an OWE). The number of OWEs is equal to the number of weights of the main MLP. The computation in each OWE can be done independently, therefore the training and relaxation phases can easily be parallelized. We report our implementation of this architecture on an Intel Paragon parallel computer and the comparison with its implementation on a sequential computer. MOTS-CLÉS : Réseaux de neurones, perceptrons, implantation parallèle, performance.

برای دانلود رایگان متن کامل این مقاله و بیش از 32 میلیون مقاله دیگر ابتدا ثبت نام کنید

ثبت نام

اگر عضو سایت هستید لطفا وارد حساب کاربری خود شوید

منابع مشابه

Navigation of a Mobile Robot Using Virtual Potential Field and Artificial Neural Network

Mobile robot navigation is one of the basic problems in robotics. In this paper, a new approach is proposed for autonomous mobile robot navigation in an unknown environment. The proposed approach is based on learning virtual parallel paths that propel the mobile robot toward the track using a multi-layer, feed-forward neural network. For training, a human operator navigates the mobile robot in ...

متن کامل

Forward kinematic analysis of planar parallel robots using a neural network-based approach optimized by machine learning

The forward kinematic problem of parallel robots is always considered as a challenge in the field of parallel robots due to the obtained nonlinear system of equations. In this paper, the forward kinematic problem of planar parallel robots in their workspace is investigated using a neural network based approach. In order to increase the accuracy of this method, the workspace of the parallel robo...

متن کامل

Kinematic Synthesis of Parallel Manipulator via Neural Network Approach

In this research, Artificial Neural Networks (ANNs) have been used as a powerful tool to solve the inverse kinematic equations of a parallel robot. For this purpose, we have developed the kinematic equations of a Tricept parallel kinematic mechanism with two rotational and one translational degrees of freedom (DoF). Using the analytical method, the inverse kinematic equations are solved for spe...

متن کامل

The efficiency of Artificial Neural Network, Neuro-Fuzzy and Multivariate Regression models for runoff and erosion simulation using rainfall simulator

1- INTRODUCTION According to the complexity of environmental factors related to erosion and runoff, correct simulation of these variables find importance under rain intensity domain of watershed areas.  Although modeling of erosion and runoff by Artificial Neural Network and Neuro-Fuzzy based on rainfall-runoff and discharge-sediment models were widely applied by researchers, scrutinizing Arti...

متن کامل

Performance of the Wavelet Transform-Neural Network Based Receiver for DPIM in Diffuse Indoor Optical Wireless Links in Presence of Artificial Light Interference

Artificial neural network (ANN) has application in communication engineering in diverse areas such as channel equalization, channel modeling, error control code because of its capability of nonlinear processing, adaptability, and parallel processing. On the other hand, wavelet transform (WT) with both the time and the frequency resolution provides the exact representation of signal in both doma...

متن کامل

ذخیره در منابع من


  با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید

عنوان ژورنال:

دوره   شماره 

صفحات  -

تاریخ انتشار 2007