Parallelization of Neural Networks Using PVM
نویسندگان
چکیده
We use Neural Networks (NN) in order to design control architectures for autonomous mobile robots. With PVM, it is possible to spawn different parts of a NN on different workstations. Specific message passing functions using PVM may be included into the NN architecture. A graphical interface helps the user spawning the NN architecture, and monitors the messages exchanged between the different subparts of the NN architecture. The message passing mechanism is efficient for real time applications. We show an example of image processing used for robot control.
منابع مشابه
Coarse-Grained Parallelization of the Advanced Neurosimulator Faun 1.0 with PVM and the Enhanced Cornered rat Game Revisited
Today artificial neural networks are very useful to solve complex dynamic games of various types, i.e., to approximate optimal strategies with sufficient accuracy. Exemplarily four synthesis approaches for the solution of zero-sum, noncooperative dynamic games are outlined and discussed. Either value function, adjoint vector components or optimal strategies can be synthesized as functions of th...
متن کاملPVM–based Training of Large Neural Architectures
In this paper a methodology for parallelizing neural network training algorithms is described, based on the parallel evaluation of the error function and gradient using the Parallel Virtual Machine (PVM). PVM is an integrated set of software tools and libraries that emulates a general–purpose, flexible, heterogeneous concurrent computing framework on interconnected computers of varied architect...
متن کاملParallelization of an Evolutionary Neural Network Optimizer Based on PVM
In this paper the parallelization of a evolutionary neural network optimizer, ENZO, is presented, that runs eeciently on a workstation-cluster as a batch program with low priority, as usually required for long running processes. Depending on the network size an evolutionary optimization can take up to several days or weeks, where the overall time required depends heavily on the machine load. To...
متن کاملParallelization of a State-of-the-Art Industrial CFD Package for Execution on Networks of Workstations and Massively Parallel Processors
Parallelizing TfC, a state-of-the-art industrial 3d CFD code for execution on both networks of workstations (NOWs) and massively parallel processors (MPPs) is one of the main objectives in the interdisciplinary research project SEMPA3. In this paper, the concept for parallelizing TfC is presented as well as results obtained with a first prototype.
متن کامل