用于手写汉字识别的分布式平台研究与实现

手写汉字识别实验样本量巨大,使得实验无法在有限时间内在单机上运行完成。而模式识别实验需要利用已有的并非为分布式实验而设计的工具,使得其利用当前的分布式中间件模型比较复杂。本文设计并实现了一种轻量级的用于模式识别实验的分布式平台——HWDC。它将网络上的多台计算资源及存储资源整合为一体,提高实验速度。使用TML对任务进行描述,以使得已有的工具可不加修改的部署到该平台上。本文研究了该平台的架构、调度策略、错误处理等。

Design and Implementation of a Distributed Platform for Handwritten Chinese Character Recognition

The amount of samples in Handwritten Chinese Characters Recognition is so huge, so that experiments are impossible to be completed within limited time on only one computer. And to utilize the current tools in pattern recognition which are not designed for distributed computing makes it difficult to do experimentbased on current distributed middleware. This article designs and implements an lighted distributed computing platform for pattern recognition experiments named HWDC. It integrates the CPUs and storages of hundreds of computers to speed up experiments. It describes experiment tasks by using TML to make existing pattern recognition tools easy to be deployed on the platform without any change. This article focuses on the architecture , messaging mechanism, scheduling strategy, and error tolerance of the platform.