Browse State-Of-The-Art

Feb. 19, 2019, 11:53 p.m. By: Kirti Bakshi


Papers with code, that is basically a clearinghouse for a number of research papers as well as their corresponding implementation code is a GitHub repository, that has been put together by Fataliyev, a Machine Learning Engineer based in Seoul.

This extensive collection, with a unique concept, has been put up with the help of various requests made by the contributors and mainly has its goals set at creating a free and open resource with Machine Learning papers, code and evaluation tables.

All the papers, besides being sorted by stars, are also arranged by year, a feature that makes it even easier to find standout research — with its corresponding code acting as an added advantage to all its users.

What is new?

The latest version of Papers With Code has just been released and as a part of this 950+ unique Machine Learning tasks, 500+ evaluation tables (with state of the art results) and 8500+ papers with their respective codes have been extracted as well. Adding on to this, the entire data-set has also been open-sourced.

Everything that is present on the site is both editable as well as versioned. The tasks and state-of-the-art data as a result turned out to be really informative to not only discover but also to compare research - and as well find some research gems that weren't known about before.

And, reaching out to a wide range of audience, anyone is free to join in annotating and discussing papers!

What is their mission?

As already mentioned before, The mission of Papers With Code is basically to create a free and open resource with Machine Learning papers, code and evaluation tables. The papers may be straight to the point, but are also pretty accurate.

The team believes that this is best done together with the community and powered by automation.

As of now, they have already automated the linking of code to papers, and are now working on automating the extraction of evaluation metrics from papers.

How can you Contribute?

Contribution made by anyone is much appreciated and if any user wants to submit a new code implementation, they can simply Search for the paper title, and then add the implementation on the paper page.

Further, if one wishes to add an evaluation table or a task, he/she can very easily see the edit buttons on the paper and task pages - all that is to be done is to just go ahead and edit!

Finally, talking about their data sources:

Most of the data come that they present comes from their very own annotation of papers. And in order to further ensure that they possess a broad coverage of Machine Learning tasks they have parsed the titles of more than 60,000 papers. In addition, the tasks and datasets have been manually annotated in 1,600 ArXiv abstracts from the last three months of 2018.

Well, The whole concept definitely turns out to be a fun way to learn about new areas of machine learning as well as staying in tune with research!

For more information regarding the same, one can go through the links mentioned below:

Source and information: Papers With Code