GCP AutoML Tables: What algorithm is used when creating a regression model?

Question

What algorithm is used in AutoML Tables when creating a regression model?
The document does not explain the details.

For example, AWS discloses the algorithm of "Amazon Forecast" service. I would like to know such information.

Answer 1

The Model Training section of AutoML Tables features and capabilities lists the various candidate model forms that AutoML Tables may search over. One of the candidate model forms listed there is AdaNet, which is itself a flavor of neural architecture search.

Recently GCP's AutoML Tables exposed the different model forms and hyperparameters evaluated during model training, per the Viewing model architecture with Stackdriver Logging docs. Note that these docs also list the candidate forms along with links to the docs or source code.

Clicking on the Models or Trials links of a particular model takes users to the portions of the training logs in StackDriver that list the model forms and hyperparameters that were tried in the AutoML job.

Unlike the Amazon Forecast service, which offers both AutoML but also lets users specify a particular algorithm, GCP's AutoML Tables does NOT allow users to choose a particular algorithm. It tries to make everything, including feature engineering, as automated as possible.