The artificial intelligence (AI) technology driven by machine learning and deep learning has gone through some amazing changes in the past decade. It is moving fast and creating many untapped business opportunities for various industries such as retail, high-tech, financial services, health care and so forth. Many enterprises are planning or prioritizing AI projects to optimize business operations and monetize the value of data. According to O’Reilly’s 2018 “The State of Machine Learning Adoption in the Enterprise” survey, more than half of the respondents are working for organizations that are evaluating AI technologies and more than a quarter have revenue-bearing AI projects in production.
When we talk about AI and machine learning, the first thing that comes to mind usually is model training. The complete machine learning lifecycle actually includes the construction of the database infrastructure, data acquisition, data preparation, parameter configuration, metrics extraction, data analysis, data training, process monitoring, hardware management and model deployment in the production phase. In fact, how to integrate AI capabilities into an application is an initial and challenging question to answer. When users select machine learning tools for their AI projects, they can face the following problems:
- Specific programming skillsets are required: traditional tools require a lot of coding by programmers. Each tool might have its own programming language associated with it. To find and hire these highly-skilled machine learning professionals or train the beginner-level developers will greatly increase a company’s labor and O&M costs.
- Mainstream AI algorithms are not supported: traditional tools support either machine learning or deep learning frameworks, which are not comprehensive.
- Workload is increased due to the complexity of the model evaluation process: to perform model comparisons and/or make parameter adjustments with traditional tools, a lot of work has to be done manually
- Model deployment is difficult: there is no standard way to move models to a production phase
AIworkflow is a unified AI management platform with a drag- and-drop modeling tool and a visual workspace to meet these challenges.
- It provides a visual, drag-and-drop interface for beginners and experts to experiment, build, train and deploy machine learning models without requiring any specific programming skills
- It provides a unified modeling platform, which integrates multiple machine learning and deep learning frameworks
- It covers the whole data modeling lifecycle, which includes data ingestion, preparation, modeling, training, testing, evaluation, deployment and prediction
- It provides a process management component, Project, to increase the reusability and portability of the modeling process
- It provides a modeling management component, Model, to support automatic model comparisons and make parameter adjustments and model deployment more convenient
Here are the main components of AIworkflow.
Dashboard: is the main interface of AIworkflow. Users need to specify a computing engine among machine learning, deep learning and tree model to start the modeling process.
Experiment: provides an interface for users to explore their data and modeling process. Any workflow can be built and any parameters can be configured to test the model design. The experiment can be edited to optimize the design. Only the information from the latest update will be saved. The model generated from this step will not be recorded.
Project: once the final modeling process is confirmed, the experiment can be converted to a project. The model created in this step will be automatically saved in the repository.
Model: a copy of the model built in the Project module will be automatically saved in the Model module and managed by the repository. Models can be compared against each other if they are in the same type under the same project. Their configuration parameters and measurements can be viewed during the comparison. The selected model will be sent to diverse deployment tools.
AIworkflow is a complete machine learning lifecycle management platform which significantly simplifies the modeling process. Model building, testing, and deployment time can be greatly reduced as well as the IT O&M cost. It helps modern corporations improve data modeling efficiency and maximize productivity and data-driven innovation, making it possible for enterprises to monetize data value in real time.