Using AutoML for AI Service Deployment

AutoML frameworks are making the life easier for users in demand for machine learning models, but without prior knowledge how to choose and train one. There are many available open-source frameworks with different working principles and performances. From the point of view of an average user, selecting and using the right framework might not be a trivial task. We elaborate on this issue and propose our solution for it. We present an idea of AutoML service that offers a selection of many open-source AutoML frameworks and provides two specific features, (1) Framework Recommendation System, which recommends suitable framework based on user’s dataset, machine learning task and time budget, and (2) Feedback Data Base, which stores data about use cases of the service. We explore how the data from Feedback Data Base can be used to further improve Framework Recommendation System, as well as the potential benefits that AutoML researchers and developers can have from this data. The features of the service are outlined in a list of requirements that will be followed in the development process.