Various research methodology tips.

Reading Academic Papers

Reading Methodology

Academic papers are not made to be read start to finish in one sitting. A better reading strategy is to read several times, with different level of attention to details each pass.

Before reading: read abstract (and skip intro/conclusion) to judge if this is relevant.

First read through: read introduction, motivation, contributions and conclusion.

Ask the following question:

• What is the research investigating?
• Why did the research investigate this?
• What was found?
• Are the findings unusual or do they support other research in the field?
• What are the implications of the results?
• What experiments could be carried out to answer any further questions?

Only read the methods, detailed experiments, etc. if I want to reproduce the experiments or the paper is extremely relevant to my research (state of the art, “competition”, etc.)

Reading List

Keep a reading list of papers that are of interest, with short explanation why I want to read them (because it is easy to forget!)

Writing

Management

It is hard/impossible to control exactly the output of a writing session. Instead of setting a content goal (“by the end of the week I will have written the introduction”) set time goals (“by the end of the week I will have spent 5 hours writing”) and stick to these goals.

Figure out most productive time of the day for focused work and set 1hr of writing each day, or set aside a single day only for writing.

Writing Tips

• Favor active phrases over passive ones for greater clarity
• Start with bullet points

Writing with multiple supervisors/co-authors

The first draft is a large chaos, akin to a wide search space, with lots of room for improvements. The first pass by a co-author, we undo a lot and do a lot of changes; this reduces the amount of possible improvements/the search space. Then a new co-author does a pass, they don’t see the first ecremage, they also introduce a lot of changes, but for them it is easier to find improvements because the search space was reduced. This again reduces the size of the search space. In the end, it should only remain a tiny kernel of search space, very dense with information and ready to burst with knowledge if you touch it.

Is there an entropy/information theory comparison to do here?