Towards Robust Incremental Learning under Ambiguous Supervision