基於服務導向架構之人工智慧決策支援機器人程序自動化

Abstract

近年來人工智慧（Artificial Intelligence，AI）浪潮席捲全球，全球各行各業皆積極地引入AI技術以提升產業價值及推動產業轉型。隨著AI產業快速地發展，對相關人才的需求也更加強烈，出現以下問題：（一）目前AI開發的入門門檻高，人才稀缺，無法滿足日益增長的商業需求。（二）AI的知識發掘（Knowledge discovery in databases，KDD）與參數調整乃至執行過程都很冗長，需要背景執行的批次處理。（三）AI模型的開發及訓練流程難以複製，不易將成功的經驗應用於其他領域。針對以上問題，我們研究將機器人程序自動化（Robotic Process Automation，RPA）引入AI模型的開發及訓練流程。（一）新增圖型化使用者介面（Graphic User Interface，GUI），將模型開發過程中的各個步驟元件化，讓AI決策支援的流程得以通過拖曳及點擊方式完成，降低開發的入門門檻。（二）提供背景執行的批次處理功能，新增度量閘道元件，可依所設定的邏輯閘道完成決策流程分流。（三）新增決策規則管理系統，使用者可儲存決策流程，未來在面對新問題時可依其資訊（如名稱、描述、資料型態、任務類別等），以AI算法搜尋相似應用的決策流程，直接匯入調整對應參數，即可達到解決相關流程知識的再利用。本系統採用服務導向架構（Service-Oriented Architecture，SOA），未來若有其他應用需求，可優先重新組合所提供的服務即可，無須重新開發。

Artificial intelligence (AI) has been going viral during the last years. People from all walks of life are introducing AI to enhance industrial value and promote industrial transformation. As this industry develops rapidly, there is a dramatic increase in the demand for talents and the following problems occurred. First, the high entry barrier of developing the AI and the existing talents are unable to meet the growing needs. Second, the processes of knowledge discovery in databases (KDD) and parameter adjustment and even the execution may turn out to be a long period of time. Therefore, we need to provide the feature of batching processing in the background. Lastly, duplicating the processes of model developing and model training is too hard to apply successfully in other fields. Measures to solve the above-mentioned problems, this study proposes a method of introducing Robotic Process Automation (RPA) into the field of AI’s process of model developing and training. By componentizing each step in the process of model development, the decision-making process can be completed by dragging and clicking which avoid repeated steps of data pre-processing and model selection on the Graphic User Interface (GUI) we provided. Next, we provide the function of batch processing in the background and add the component of metrics gateway. Users can complete the route of decision-making process that based on the logical gateway that had set up. Finally, a new decision-making rule management system is added that users can store decision-making processes. When facing new problems, they can search for decision-making processes of similar applications with AI’s algorithms based on their information (such as name, description, data type, task category, etc.). At the same time, users can import the decision-making processes and adjust the corresponding parameters directly to achieve the reuse of the knowledge process. This system chooses to use Service-Oriented Architecture (SOA). If there is any demand for other applications in the future, all we need to do is reassemble the features we provided instead of redevelopment.