If I have seen further, it is by standing on the shoulders of Giants— Isaac Newton.
The above quote first made its appearance in a letter that Newton wrote to Robert Hooke in 1675. Since then, the statement has been used to symbolize scientific progress and advancements. It is often said that one of the best ways to keep up to date with the latest happenings in the field of machine learning is by reading research papers. However, this is easier said than done. Reading research papers is not everyone’s cup of tea. While many find them intimidating, others find it impossible to keep up with the daily dose of published papers. Arxiv — one of the go-to sites for finding such papers has over 1.9 Million submitted papers( as of October 2021) and more being added every day. To make things worse, there is a `fear of missing out’ on the latest in the field. If you fall in this group, do not worry, you are not the only one. Many people are sailing in the same boat, so much so that there is a complete lecture by Andrew Ng on reading research papers.
Reading research papers is truly an art that can be developed over time, starting with some handy tools. In this article, I’d like to share a few such tools that I use to organize my favorite research papers and also get up to date with the latest ones. This isn’t an exhaustive list but is a good starting point for those new to the area of research and literature overview.
1. ArXiv Sanity Preserver
I use Arxiv Sanity Preserver to literally preserve my sanity when handling research papers on ArXiv. A site created by Andrej Karpathy, Arxiv Sanity Preserves, is a web interface to help you find your favorite papers and what is trending in the field. The site provides you with a search engine to find papers on any topic. You can then save your favorite papers in your library and access them later. Based on your searches on the site, the site will also provide recommendations that will improve over time. Here is a great intro video by Karpathy which explains how to navigate the site.
The current site, according to the author, is on life support, and he is working on version 2.
Fun Fact: Badmephisto – famous for his CFOP Speedcubing tutorial videos on YouTube is actually Andrej Karpathy 😲
2. ArXiv Vanity
Arxiv Vanity is a helpful web interface for viewing research papers from arXiv. That is to say; it renders an arXiv paper in a more readable format that is pleasing to the eye. As per the authors:
arXiv Vanity renders academic papers from arXiv as responsive web pages so you don’t have to squint at a PDF.
3. Connected Papers
A research paper is not born in isolation. On the contrary, every paper is connected to several other papers. A reader would be interested in exploring in references and citations mentioned in the paper. Connected papers is a visual tool created to address this specific point. It creates a graph consisting of the papers having the strongest connections to the paper of interest. In addition to this, it also showcases both prior and derivative works related to the paper.
4. Mendeley Reference Manager
I first learn about Mendeley Reference Manager in a forum. Mendeley reference manager is a free tool that helps you store, organize, and annotate your research papers in a single place. The notebook facility enables you to make notes while you read and also collate various highlights that you make in your paper. The demo gif below should give you an excellent idea regarding its usage.
5. Papers with Code – Methods Corpus
We all love the site Papers with Code– a free and open resource with Machine Learning papers, code, datasets, methods, and evaluation tables. However, I would like to point to the Methods section on the site, which I find particularly helpful.
This section contains well organised and category-wise curated papers. Each category further includes papers stacked by the type of method used in them. For instance, there is a Computer Vision category containing papers related to computer vision techniques. It is then further divided into 853 subcategories, each containing a collection of papers based on a distinct computer vision method.
Conclusion
The tools and sites I have mentioned in the article above have helped me immensely. I now make it a point to take notes while reading papers and then use the notes to create articles and posts. The field of machine learning is changing rapidly, and new research and state-of-the-art implementations are coming out every other day. However, such progress has been due to the unyielding work and knowledge of those who came before. Therefore, reading research papers, understanding, implementing, and building upon them is the real essence of research and scientific progress which brings us back to Newton’s famous quote – "If I have seen further, it is by standing on the shoulders of Giants."
