论文修改意见 (ZengWX 20211028 伪代码小问题)

1. The high-level instances $\mathcal{H}$
   $\to$ 同位语
   The high-level instance set $\mathcal{H}$

2. The IP.
   $\to$ 在空间足够的时候可以不用缩写

3. $k=r_c$; // $k$ is the number of instance prototypes of each bag.
   $\to$ 注释里面不一定完整的句子
   $k=r_c$; // The number of instance prototypes of each bag.

4. Compute the representativeness score
   $\to$ 代表性已经是一个指标
   Compute the representativeness

5. Merge the top-k score instances into C;
   $\to$ Merge … into 有点怪
   ${\mathcal{C}}^{\prime }$ = the set of top-k representative instances
   $\mathcal{C}=\mathcal{C}\cup {\mathcal{C}}^{\prime }$

6. // The HI.
   $\to$ 前面已经说明它是什么
   // Initialize HI.

7. Merge the top-nh score prototypes
   $\to$ 用什么来衡量 top?
   和前面保持风格/句式一致

8. // Step 2. Execute bag embedding.
   $\to$
   // Step 2. Bag embedding.

9. Train a single-instance classifier F(\cdot) with the vector-label pairs (V_i, y_i);
   $\to$ 应该是新空间的一个数据集
   Train a classifier F(\cdot) with the new dataset {(V_i, y_i)}_{i=1}^n

10. Output the final classifier F(\cdot) and the high-level instances H;
    $\to$
    Output F and H;