Want to know about selecting papers for the research seminar? I've tried to compile a few resources and summarised points from the introductory seminar by our prof to help you with the process.
A fair warning, these are only pointers. There are no absolute answers, so be prepared for exploring and lots of googling.
Disclaimer: If you are into research, save me the embarrassment and do not read a word further. This article is not for you!
First things first, how do I select a research paper of my interest?
A good heuristic to evaluate the value of a paper is to see if you
1) Understand the title
2) Like the title.
If you have absolutely 0 experience in any field, then you are a free bird. Just pick one at random and go through it.
You could follow one of the multiple routes
Route 1:
It is mandatory that you present a paper from a Q1/Q2 journal or a top conference such as :
1) IEEE journals (not conferences)
2) IET
3) All IEEE transactions
4) ACM journals
5) Springer journals (not lecture series)
6) Elsevier journals
Conferences:
1) CVPR (Computer Vision)
2) NIPS (Machine Learning and Computational Neuroscience)
3)ICML (Machine Learning)
4)ICDM (Data Mining)
5)KDD (data science, data mining, knowledge discovery, large-scale data analytics, and big data.)
6)ECCV (Computer Vision)
7)ICCV (Computer Vision)
Note that this list is NOT exhaustive and excludes a lot of top tier conferences.
Through our Manipal Student ID, we have free access to papers on the IEEE website as well as several other journals/sites like Science Direct, Research Gate, etc. You could visit the websites ->type keywords ->look for titles that interest you.
You could directly go through, say, 'ECCV 2018 papers' and read a few abstracts to shortlist 2-3 papers of your interest (An abstract lets you get the gist or essence of your paper or article quickly, in order to decide whether to read the full paper).
Route 2 (personally recommend this one -100X more fun):
Instead of aimlessly typing keywords on google you could go through blogs and GitHub repositories that compile the state of the art research papers.
Here are a few examples of how these blogs and Github repositories look like:
https://ai.googleblog.com/: You won't regret opening the link I bet you.
https://paperswithcode.com/: is a unique and useful resource that presents trending ML research along with the code to implement it.
https://towardsdatascience.com/latest-computer-vision-trends-from-cvpr-2019-c07806dd570b: Here the medium author has compiled a list of her top 5 favorite research papers from CVPR 2019. An advantage to medium articles is that they mostly summarise (aka explain the intuition behind) the papers in layman terms.
http://nlpprogress.com/: This is a portal that tracks progress in various NLP subfields along with the latest papers, code (if it's available).
https://github.com/JackieZhangdx/WeakSupervisedSegmentationList: An example of a GitHub repository compiling the latest 'State of the art' in a specific domain - 'Weakly supervised segmentation'
If you like a research paper and wish to look at similar other papers, the easiest way to do so is in the references of the paper (mentioned at the end of the paper). So take the best paper, from the above websites, dig into the references.
ROUTE 3 - YouTube:
You could also go through YouTube channels like 'Two Minute Papers' for paper summarisations or for conferences that post snippets or entire videos of the paper presentation.
Two of my personal favorites are :
Two-minute papers, Károly Zsolnai-Fehér : My personal favorite- his videos are a treat. The explanations and demonstrations are so simple that anyone without machine learning knowledge can still take away the gist of it. Here is a recent example.
Henry AI labs: He summarises new papers that research labs from companies such as Google and Facebook are publishing.
If you wish to explore more complex paper analyses (of AI papers) then do checkout Yannic Kilcher and Leo Isikdogan. 3blue1brown, Amii intelligence, simons institute are few others not restricted to AI.
FAQs:
1) Research is not my thing, (I want to focus on placement prep) how to select an easy paper so I can just be done with it?
A- There is no such thing as an 'easy paper'. (If math is not your strong suit, you definitely might want to go through the paper and pick something with fewer 'weird notations' and symbols, although a certain amount of math is unavoidable in any 'good paper'). Once you have selected a paper, watching the paper presentation video on YouTube and/or reading up related Medium articles can make this task a lot less arduous.
2) Resources on how to summarise and ensure it isn't a mere copy of the original paper?
The final report format specified is very similar to a research paper in itself: Final Report format
3) Can you pick survey papers or review papers?
Survey papers or review papers summarize and organize recent research results to integrate multiple studies and adds understanding to work in the field. Picking a survey or research paper will not allow you to adhere to the report format as discussed by the faculty i.e, you will not be able to answer questions like 'What methodology has the author used', 'How is it different from the existing work' etc.
4) If I know I like a given field, say -cloud computing, how do I find papers in that field that will meet the criteria specified by the department?
One way is to look for the top international conferences or journals in that field. These top conferences and journals will most likely be accepted for the paper presentation by the department. Then look at the 2018 or 2019 list of papers in that conference/journal. You might want to use sci-hub(a website that provides free access to millions of research papers and books) if you can't find the PDF version of the paper elsewhere. You can also find links to blogs like this one where authors pick the top 10 papers from a conference and summarise them.
Few important points to keep in mind (as explained by a CSE faculty):
1) Write the summary in your own words. Do not directly copy-paste sentences from the base paper. You are expected to perform a plagiarism check and attach the results at the end of your report.
2) Don't have hanging figures or tables i.e, do not insert an image or any graphic in your report without referencing it in the text.
3) Use Scimago or any other tool to evaluate the quality of the journal - https://www.scimagojr.com/journalrank.php
4) Attendance will be taken during the seminars.
5) 20 minutes to present + 5 mins for classmates to ask questions + 5 mins for professor's QnA
6) After shortlisting your base paper, you are expected to get your paper approved by your assigned mentor. Then, keep your mentor updated with the progress of your work.
PPT and presentation guidelines :
This was suggested to section A, batch A1- CSE. You might want to cross-check with your faculty for the respective guidelines:
Minimum 15 slides (excluding the first)
Don’t put too much content. Max 2-3 lines on one slide
The first slide is a vague introduction of the paper and an outline of what the paper is all about
The second slide will be the literature survey. What the author intended to do (aim), the methodology followed, what were the results, and the drawbacks of each (4-5 lines/ papers max )
In the third section, your goal is to identify the research gap (or knowledge gap in the literature survey. A research gap in layman terms is something that has not yet been explored or is under-explored. This could be a population or sample (size, type, location, etc.), research method, data collection and/or analysis, or other research variables or conditions. Proceed to talk about how the base paper is trying to address the gap. What are the objectives of the base paper aka the hypothesis?
Explain the methodology followed in the base paper with the help of algorithms, flow charts, etc.
Elaborate on the results (tables/graphs) and what the results have been compared with (aka metrics or benchmarks).
Most important - The section on future work: your observations and what could be incorporated i.e, a section on the research gap (what you think could improve upon the results or methodology), if any, in your base paper.
Please do let me know if I have missed out on anything (particularly w.r.t. topic selection) or if there's any additional content you would like to see in the FAQs section.
Hope it was useful :)