Hello, I have a big problem when I train my CNN model for binary classification. The training loss decreases but the training accuracy also decreases. When I change the learning rate, the training loss is blocked and the training accuracy increases very sharply. Does anyone have any ideas about this problem? Best regards!