Next Generation Protection machine learning algorithms

Hello Community,

I did not find much information on the documentation about machine learning algorithms and models used in Microsoft Defender ATP. Therefore, I would like to ask the following questions:

• Which is the data that gets ingested to train the models?
• Which is the frequency of the periodic re-training of the models with current data?
• How long the previously used data to train the models impacts on the decision?

Thank you in advance.