SCI thesis writing

The writing of the thesis is divided into four orders ： Reading papers → Identify innovations →Coding／ experiment → Thesis writing .

1、 Reading papers
The premise of publishing papers is to read a lot of papers ！！！ Literature reading is divided into three stages , Find the direction at the beginning , Medium term key breakthroughs , In the later stage, he dabbled extensively .

Reading papers at the initial stage requires intensive reading word by word , The direction need not be strictly limited , You can read all the paper chains involved in the papers you are interested in . A paper takes one day , English papers + Chinese sharing , Number of papers read in the early stage 30 Article above , It can improve the reading ability of academic English and the accumulation of professional terms .

During the mid-term reading of papers, we should focus on intensive reading , Strictly limit the research direction and the paper chain involved . The time of key papers should be controlled in half a day , A general paper is an hour , Repeat key papers + Source code learning , The best number of reading papers is 10 Article above . Understand the evolution of learning technology 、 Learning method innovation and sorting innovation method chain .

A few intensive reading in the later stage + Extensive reading , No direction , Focus on your own direction + Dabbling direction .

Focus on the paper for two hours , General paper for half an hour , Follow the latest development of research , Understand the general progress in other directions , Think about the introduction and grafting of innovative points .

2、 Identify innovations

You can determine the innovation of your paper from the following four aspects ：

1. Changes to data sets ： noise 、 Geometric transformation 、 Occlusion 、 Light conditions 、 Scenario dependency

2. The problem with the model ： Model volume 、 Speed of reasoning 、 Convergence difficulty 、 Non end to end 、 Post processing optimization

3. Structural replacement ：transformer、FCN、AE、

4. Application of specific scenarios ： The general model considers generalization ability — Consider specificity for specific applications . Such as night detection 、 Underwater detection 、 Fish eye camera detection .

The other is to remember A+B+C/2.5 The laws of

A： The inherited innovation of this research direction （ Natural evolution ）

B： Existing methods in other directions （ Grafted to other tasks ）

C： Innovation in detail （ Data to enhance / Data sets / Loss function design ）

For example, the following article CVPR2021: CutPaste, The use is A+B+C/2.5 The laws of .

A： Natural evolution cutout—cutpaste+B： Existing methods Self supervised learning pretask For anomaly detection +C： Detail innovation Gaussian probability density estimation （GDE） Abnormal judgment

3、Coding/ experiment

principle ：1 The code of this paper is reproduced （ At least understand the code implementation ）>> Run through multiple projects demo

1. find baseline The code of the paper ;

2. stay baseline Minimize the implementation of the desired function on the code ;

3. Gradually realize the final functional code , At the same time, the implementation verifies the effect of each part of the design .

4、 Thesis writing

01： Writing strategies ：

choice 2 About model essay , To analyze the structure of the paper （Introduction）、 Key words and sentences （Related Work）、 Language style （Method）、 Experimental design （Experiment）、 Drawing style （Conclusion） And story design （References）.

02： Thesis writing skills

（1） Thesis writing skills —— Pay attention to the title

use ⼀ In a word, summarize the work you have done

Consider the impact of search engines , Include keywords

It can be novel

（2） Thesis writing skills —— Home page plus pictures

3） Thesis writing skills —Introduction Direct column contribution

There is no need to introduce how the various parts are organized ;

Directly speaking, what contributions have been made ;

Indicate the contribution position .

Source of the article ： Deep learning beginners