Andrew I want to understand how dropout behaves during the testing or inference phase. Why are neurons dropped during training but not during testing? Can someone also explain how predictions remain accurate after dropout is removed?