Advanced NLU v/s Standard NLU

[saurav_nair]

Can someone help me understand when should we use Advanced NLU in Dialogflow CX?

From my understanding, standard NLU is suitable during development as it has the auto-train ability and once we deploy the agent to Prod (let's assume its a different agent and not an environment), we can enable Advanced NLU as we don't expect any changes directly made to the prod agent. 

But if we do this can we guarantee that the testing done for the dev agent with standard NLU is sufficient and there won't be any unwanted behavior?

1. Is Advanced NLU recommended for all the agents built in Dialogflow CX or is there a specific criterion for choosing the NLU option?
2. What are actual ML algorithms used to train in Standard v/s Advanced NLU configuration?