How to Do Research Meta Discussion
Published:
These are the notes from a lab meta discussion on doing research, reading papers, and finding ideas. The ideas are presented in no particular order.
Assignment
- Read 3 papers
- take notes about how you read the papers
- one enjoyable paper
- different knowledge levels for each
Questions
- Why are you reading that research paper?
- open problems
- related work
- advisor told me
- to implement
- to learn about a field
- lit search
- maintaining knowledge
- trends in a field
- interest after conference
- contains answers to homework
- seminal papers
- How does one pick a paper to read?
- number citations
- by reference
- by google scholar
- by author
- by conference
- To what level of detail should I read?
- 3 passes (from paper)
- introduction, background, related work useful if unfamiliar
- levels depending on knowledge of area
- find “these are our contrabutions” section
- Is this paper worth reading?
- too many papers to read everything
Approaches
- Write why reading to check if you succeeded in reading for your goal
- Be able to explain paper to someone else
- loose
- depending on your goal
- Read blogs, Medium articles for high-level entry points
- take a class
- read books
- Surveys more useful when you are familiar with a field (truth in this?)
- Find your comfort level for entry point
- Videos at 1.5 speed
- Don’t spend all time reading papers
- Recognize when to stop reading a paper
- Don’t always seek to understand every minor detail of a paper
- Recognize what method works for learning a paper or topic
- 4 or 5 hours of deep work per day, use it wisely
- Explore the network of papers
- Read multiple papers at once
- active compare
- Paired reading
- Have a [good] pen in your hand
- Take breaks
- 3 passes
- title, abstract, introduction, and conclusion
- key points, figures, references, problem and solution
- fine details, implementation; be able to re-create
- Watch the talk of the paper
- Authors may emphasize certain parts more than others
- Look at the paper’s code
- Build related work before writing your paper
- Use a whiteboard
Tools
- Reference manager (stores pdf)
- Mendeley
- Zotero
- Powerpoint
- gnuplot
matplotlib
Reading Critically and Creatively
- Wear the skepticism hat, read papers thinking What is wrong with this?
- Wear the creativity hat, think How would I improve this work?
How to write notes
- low-level notes; write while reading
- high-level notes; 2 sentence summary afterwards