【论文写作】9.15 修改心得

本次修改主要部分为摘要。

一、关于论文结构

1、how-what 的写作思路（怎么做与优点）。
不仅仅是表意清楚，较为定式的写法在能力不足以炫技时是一个比较稳妥的方法。

Finally, we introduce a focal frequency loss in network training to improve the recovery of high-frequency details. 
This loss dynamically focuses on difficult frequencies by down-weighting easy frequencies.
->
Finally, we introduce a focal frequency loss in network to focuse on difficult frequencies by down-weighting easy frequencies.
It effectively improves the recovery of high-frequency details.

2、论文有树型结构，每一句都是一个节点。

3、挖的坑后面有没有填？对自己挖的坑负责。
每一个词，都要对其作出相应的解释，而不是一个随便就跳过去的词语。

It learns stratigraphic features from seismic data without the need for complex initial estimation and manual intervention.
-> stratographic features

对这里提出的地层特征，如果采用，那么后文中需要对这一点进行一定程度上的相关阐述。

二、关于遣词造句

1、句子长度
保持在15-18个词一句即可，要留心其信息熵，尽可能的使一句话包含较多的有用信息。这一点其实与 “句子具体” 是相辅相成的。
“更多的干活，更容易理解”——老板如是说。
比如，怎么个 difference：

It effectively extracts edge details in seismic data by calculating the difference between pixels.
->
It effectively extracts edge details by calculating the difference between pixel pairs along different directions.

First, we design a novel oblique difference convolution inspired by an existing center difference convolution.

但同时，也要追求一定程度上的简洁，不必要的形容词不要一股脑往上加。这一点其实与日常对话也有相似性，现实情况中从句是一种效率比较底下的交流方式，甚至不如单独的 n 修 n。

Experiments on SEG salt datasets and four synthetic datasets from OpenFWI
-> Experiments on SEG salt and OpenFWI dataset

2、具体
能够具体的描述（如数量），就不要使用模糊的词语。

difference convolution -> these two types of convolutions

包括专业术语。

In this paper, we propose a network...

是怎样的 network，若不做解释，很可能会理解成最底层的神经网络，而事实上，采用的网络显然要高级的多，这里最好直接点名。

3、句子内部的逻辑
如果一句话会存在理解歧义，那么应该添加对应的限定词来避免这种情况。

It learns stratigraphic features from seismic data without the need for complex initial estimation and manual intervention.
% It learns stratigraphic features from seismic data without requiring complex initial estimation and manual intervention.
% It learns stratigraphic features from seismic data without complex initial estimation and manual intervention.
% Without complex initial estimation and manual intervention, it learns stratigraphic features from seismic data.

对于原句中的 `the need for'，通过这一短语明确表明后者与前者的关系，若采用类似改句2的写法，可能会导致读者理解歧义，对两者关系产生不必要的误解。同样的，改句1和改句3能起到相同的表意。
另外，还需要注意名词之间的结构关系。原句中显然不太符合 combine 的关系。

Second, we combine depthwise separable convolution structure with difference convolution to obtain a lightweight and effective convolution module.
->
Second, we incorporate these two types of convolutions respectively into the depthwise separable structure to obtain two lightweight modules.

4、避免同一个词语多次出现
平替的一大原则，仍然是：尽可能的使一句话包含较多的有用信息。

5、更加地道，更加通顺
这一点需要日积月累，平时的注意慢慢改善，尤其是要注意从小到大被灌输的固定句式，这一点分情况要小心。

DC-Net: A Full Waveform Inversion Method Based on Difference Convolution
-> DC-Net: Full Waveform Inversion with Difference Convolution

have room for improvement 的表意，与单个形容词对比。

However, popular DL-FWI networks have room for improvement in recovering detailed edge features and high-frequency details in seismic data.
->
However, popular DL-FWI networks are ineffective in recovering detailed edge features and high-frequency details.

句中信息的顺序也是需要考虑的一个点。