Pregunta de entrevista de AnyVision

how to handle an unbalanced dataset? Corrupted Labels? Loss Function, L1 L2? Batch Norm?