Leveraging Google ML for hosting Machine Learning models

Presenters: Trygve Karper and Erlend Vollset, Cognite AS.

At Cognite, we develop an industrial dataplatform for retrieving and connecting all the various data stored
in the jungle of data systems belonging to our clients. Typically, sensor data are stored in one system, maintenance logs
in another, equipment overviews in a third, technical documentation in a forth, and so on. We expose this data to application developers,
and in particular, to Data Scientist through a REST API, with all data connected to a common data model.

Our customers are currently using several consultancies and specialist companies to develop Data Science models
using data coming from the platform, and writing the results back to the platform. In order to avoid managing to many
hosting solutions for these models, Cognite also offers an hosting environment where these models can live.

In this talk, we will take a deep dive into what Google ML is offering, what is not (yet) offering, and how to make it suitable
also for offering hosting to third party developers. Moreover, we will show some concrete models for time series data
and how those models are being deployed in our hosting environment.

PS: pizza and softdrink will be provided