paper writting

well-organized: storyline -> abstract and introduction -> methods -> related work -> experiment

1. Storyline

   • Make sure that the storyline is reasonable and fluent before your writting.

2. Abstract & introduction:

   • Based on the refined storyline, to form abstract and introduction, a extension of abstract.

3. Related work

   • It related to our selected baselines, so I think it belong to the survey work before this research project, and to update it when you read related papers.
   • Another important point is to divide the sub-fields of your current research topic, you can find related works accurately and comprehensively.

4. Methods

   • To fully understand the proposed method, generally the method is inspired by other papers or books.
   • It’s important to know about what the method is and how it can deal with your problem.
   • More importantly, forming core combined method based on those refered algorithm can help us to explain our story line in therotical analysis.

5. Experiment

   • Comparion is important and classify those baselines into serveral group based on its relateness.
   • To ablate the role of parameters in experiments and introduce ‘special case’ to outline the effect of the proposed method.

well-written: abstract -> introduction -> method -> experiment -> conclusion

1. Abstract

   • Describing the scenario of your reserch problem.
   • The shortcoming of the existing methods.
   • Our inspiration, which means what sub-field we selected as research point for this prolem.
   • What problem we will encounter in using this sub-field method, eg.. 1 …, 2…
   • To highly summarize how our approach to solve thos problems.
   • To brief describe the experimental results.

2. Introduction

   • From large research point (MTL) to sub-field research point (MTL with disjoint datasets).
   • Describing this scenario and the disadvantage of existing methods.
   • Our method, explain it.
   • Concluding contributios.

3. Related work

   • Directly related work.
   • Methods related works.

4. Method

   • To describe the problem in mathmatical method (inputs [images and labels], trainable parameters).
   • Inspiration work
   • Hierarchical method description, for example, data selection(confident [confidence score, local density], distribution), label vector interpolation.

1. Experiment

   • Datasets: split rule (make sure before experiment).
   • Comparison: directly related to the part of related work, brief how you reproduce those baseline in your prpblem scenario. and then to analyze the result to outligh the advantage of the proposed method.
   • Ablation: explaining the introduced hyper-parameter, surely, it’s necessary to classify those ‘special case’ to facilate the description.
   • Considering the qualitative analysis if possible.

2. Conclusion

   • Highly conclude what the proposed method and how it deal with problem.
   • The demonstration of the experimental results.

1. Paper organization: overall organization and the organization of each part.

2. Hierarchical description in each part: logical description in each paragragh.

3. Expressions of mathematical formulas: unification of superscript and subscript in formula, and refining it during writting.