Good Citizens of Research Community
On average, I am on the road for 2 hours every day. I have much time commuting to and fro my workplace. Thanks to the infamous traffic of Bangalore. Instead of cribbing (trust me, I did a lot of that) I decided to put my time to good use. I listen to podcasts, sleep (sleeping is a good use of time), watch videos with less visual content and more informational audio, and so on. This blog post is also a result of my yearning to make good use of my travel time.
In my quest to find informational content, I stumbled across videos of the premier Computer Vision Conference Workshop CVPR 2018 — Good Citizen of CVPR. This workshop is a one-of-a-kind initiative to help the research community, especially novice researchers. The line-up consisted of some of the giants and distinguished scholars in the field. I found the talk relevant to all researchers, irrespective of the area of research. Here is an attempt to summarise it. I hope you like it as much as I enjoyed the videos.
Note: Being a junior researcher, I have outlined points which I think are relevant. I would like to refer the readers to the original videos for some more pointers and great talks.
As rightly put in the opening remarks, non-technical factors are the things that we care about the least. Somewhere between the last-minute experiments and the paper drafts, it gets lost. I opine that these talks can act as reference material and guideline.
What does it take to write a good research paper?
I highly recommend the content by Bill Freeman, here and here. I particularly like this graph from the first talk.
Always write good, creative and original papers which have a significant impact on your career. Nothing even remotely less.
In my opinion, this applies to anything we write. Be it blog posts, reports, or white papers. Creative and original work always gains traction.
Paper organisation is one of the significant non-technical factors that we tend to ignore. The below slide gives a concrete outline of the same. The main idea here is to make the paper an easy-read for anyone interested.
- The speaker stresses on a dynamite introduction. What is the paper about, what problem it solves and why that problem. This point applies to abstract as well to make a compelling case to the reader and pike the interest to read the whole article/paper.
- One teaser figure which encapsulates the idea we are trying to convey.
- Underutilized technique: explain the main idea with a simple, toy example.
- The paper should clearly describe the main idea with ample experiments and factual results in the experiments section. The conclusion should not repeat what’s already said but address what the paper has opened up.
The speaker feels that future work is a weak way to end a paper. It would be like conveying the things that could not be done before the deadline. I’m guilty of doing this too. It seems like an easy way to end a paper. Maybe if written subtly, it could be a good ending. Would you end a blog post with future work? 🤔 Let me know your thoughts in the comments section.
General writing tips:
- Keep the reader in mind. Treat the reader like a guest in your house.
- Be succinct. Omit needless words.
- The figures and captions can help tell a story.
- Have a kind and gracious tone. Be clear and reliable.
The talk also captures some interesting and informative quotes from various researchers. Do check them out!
Research Culture
The next few researchers talk about the research culture as a whole and particularly in academia and conferences. They talk about how to be welcoming and make sure all the attendees and especially the newbies don’t feel out of place.
I’m particularly interested in this slide from the talk by Prof. Sven Dickinson.
I can relate to all the points mentioned here. Research, especially in the field of Artificial Intelligence and Machine Learning is overwhelming. There is so much of new ideas and papers coming out that you feel you are sinking in it. I feel both excited and intimidated at the same time.
The speaker mentions how he coped with it with help from his mentors and role models. The main take away from this slide is that everyone experiences this at one point or other and with help, all these can be overcome.
Another resource that I like for beginner researchers is this essay from John Schulman of OpenAI. Do check it out. I like the insights about progress via continual review and aiming towards a larger goal.
Incremental steps should be moving you towards a larger goal that represents a non-incremental advance.
It is difficult to keep updated with the field especially when a plethora of papers are getting published. Twitter is a good resource to keep updated with new and upcoming research. You can also find hidden gems if following the right people. I also listen to Daily Arxiv Radiostation to keep updated with papers. It’s an automated podcast which reads the abstracts of top 10% papers in each field.
How to give a clear talk?
The speaker starts with the point of understanding your audience. The audience plays an important role in determining the direction of the talk. Are they a broad set of people or narrow or specific. The level of detail and use of the language/grammar/jargon depends on this. Rephrasing and reiterating the main take-home message is necessary to hold the attention. A one-liner or punchline for the result is a good way to make sure it is etched in everyone’s mind.
Be concise. Though this point depends largely on the time allocated for the presentation, we need not put out all the details. Be it while talking or on the slides. Deciding what to omit forces us to deliver an important message to the audience. It is important to remember that the main aim of the talk must be to pike the interest of the audience on the field or topic or paper. If done successfully, they will go back and find out more about it. Shorter text phrase in the talk and on the slides is the way to go. Technical slides can especially become too wordy. Try to reduce the cognitive load.
These points perfectly suit for the prep of an elevator speech as well. An elevator speech is a quick synopsis very essential for networking in conferences, meet-ups, events etc., Short. Quick. Concise. It is always good to have one ready at the back of the mind. Never know when it might come in handy.
My main takeaway here is the flow of slides. I always think of what to say for that particular slide in the talk and don’t explicitly think about leading the current one to the next. The speaker here makes an interesting point of thinking about where did we come into the slide from and where do we lead to. This helps in taking the audience through your chain of thought. Stage fear and being nervous is common.
I have tried my best to summarize the main takeaways from the tutorial. I have missed many other great talks like Principles to Thrive in the Research Community, Calendar. Not to-do lists. as I don’t want to bore you with a very lengthy post. But I recommend listening to all talks in leisure. The last talk also gives a brief insight into the world of reviewers and might be a good starting point to understand how research papers get reviewed, accepted etc.
There was another workshop at CVPR 2020 on How to write a good review? It also includes Bill Freeman’s talk on how to write a good paper. It is always interesting to see such workshops at premier conferences. Looking forward to watching it!