Getting worse results when binary cross-entropy loss is computed from logits in Keras
My model (Keras framework) performs significantly worse when the loss is calculated from logits:
My model (Keras framework) performs significantly worse when the loss is calculated from logits: