And then we use another prediction network on just the extracted explanations for optimizing the task performance.
確定! 回上一頁