智能数据管理系统 (ART/233CP)

智能数据管理系统 (ART/233CP)

  • 智能数据管理系统 (ART/233CP)
    ART/233CP
    平台
    07 / 07 / 2017 - 06 / 05 / 2019
    11,765

    刘文建博士

    The deliverables comprise of (1)Developer Services, (2) Machine Learning and (3) Distributed Systems. 1. Developer Service: 1a. Research and design documentation for ML web framework, to cater for Chinese handwriting paper form and document management system used by different organizations 1b. Web based ML setup portal for cloud and distributed devices, for easier backend and frontend deployment for paper form processing and increase the commercialization chance 1c. Data Visualization modules, for report generation and visualization of data flow, training status, model definition, knowledge tree and accuracy, enable one to set strategy and improve accuracy for specified context and industrial applications. 1d. Sample code and projects with App and Web, and create the building block for Chinese handwriting paper form processing, increase the commercialization chance 2. Machine Learning: 2a. Research and design documentation for ML models and algorithms , to represent ML checkpoint in timeline and branches, with incremental learning from sample collected through distributed system, and cater for Chinese recognition in specific context and domain semantics 2b. Detail design of models and algorithms, to improve accuracy of Chinese recognition in specific context and domain semantics, create classifiers for different parts of form and fields of document. 2c. Development of Models and Algorithms, including the heterogeneous software in various platforms with consideration of distributed systems. 2d. Integrate previous ML engines from seed projects for Chinese Handwriting Recognition and Signature, also provide a framework for future integration of multiple ML engines and OCR for non-Chinese. 3. Distributed Systems: 3a. Design documentation of core engine/framework, include the enhancement of ML engine for knowledge topology and handle Chinese handwriting image in different context 3b. Data and Machine Learning frameworks integration, for incremental machine learning to handle more Chinese handwriting samples from large number of frontend devices and users 3c. Devices management system, for device to upgrade automatically with centralized management, and allow human intelligence to involve verification and correction to generate more training sample and feedback. 3d. Agents management system, for collection of labeled samples in different context and consider the knowledge topology from large number of frontend devices and users. 4. CS deliverable: (for Broadlearning) 4a. Reference design of document and database management system, with Chinese handwriting support for financial, education and other industries, including the frontend in various portable devices and computers, and backend in cloud or dedicated servers. The reference design also provides SDK and interface with commercial system, it will handle at least 3 kinds of paper forms: (1) change of address, (2) application of account and (3) termination of account.The deliverables comprise of (1)Developer Services, (2) Machine Learning and (3) Distributed Systems. 1. Developer Service: 1a. Research and design documentation for ML web framework, to cater for Chinese handwriting paper form and document management system used by different organizations 1b. Web based ML setup portal for cloud and distributed devices, for easier backend and frontend deployment for paper form processing and increase the commercialization chance 1c. Data Visualization modules, for report generation and visualization of data flow, training status, model definition, knowledge tree and accuracy, enable one to set strategy and improve accuracy for specified context and industrial applications. 1d. Sample code and projects with App and Web, and create the building block for Chinese handwriting paper form processing, increase the commercialization chance 2. Machine Learning: 2a. Research and design documentation for ML models and algorithms , to represent ML checkpoint in timeline and branches,

    博文教育(亚洲) 有限公司
    京信通信有限公司
    创次元有限公司
    柯尼卡美能达商业系统(香港)有限公司


    今天,大多数中文手写图像识別引擎是相对静態不变的,即使是相同的中文手写,用户也需要重复修正 。此外,公司使用的设备(智能手机,平板电脑,桌上电脑和笔记本电脑)种类繁多。 在过住的应科院种子项目中,我们开发了一概念系统,应用机器学习(ML)在中文手写图像识別,吸引了一些本地公司,支持开发这个平台项目,希望进一步开发更有效率地处理纸文件之技术。 在这个平台项目“智能数据管理系统”(IDMS)中,將通过人类智能(HI)与分佈式系统的交互,开发具有不断进步之中文手写图像识的文档处理系统,使用家之验证结果於不同的前端系统成为ML的新培训样本,提高中文手写识別的准確性。该平台具有可扩展性和灵活性,適用於具少量开发人员的中小企业, 我们更提供开放之软件开发工具包(SDK)和范例原代码,使用较少资源也可开发具工作流程的应用程序。我们的框架还通过知识拓扑来改进ML,实现不同领域的准確结果。 此外,该平台的技术和框架可以进一步应用到其他行业与其他资料来源,而资料產生可来自不同的智慧设备,如医疗保健和智慧城市地区。