One researcher's mission to make machine learning research reproducible

One of the key problems of machine learning research is the “reproducibility crisis,” where researchers struggle to reproduce the results presented in ML papers. Scientists who publish papers often omit to publish their code and data, which makes it hard for others to reproduce their techniques and use them for future research.

One researcher who was frustrated after weeks of trying to reproduce a certain paper took matters into his/her own hands and created a website called “Papers Without Code,” which will publish a list of machine learning papers that don’t have a source code available and whose results are unreproducible.

TechTalks spoke to this researcher, who goes by the username ContributionSecure14 on Reddit, about the problem with unreproducible machine learning papers and how Papers Without Code can help deal with this problem.

Papers Without Code is not aimed at shaming researchers who don’t publish their source code and data. On the contrary, it is meant to help promote a culture of reproducibility in the machine learning community by creating a hub where researchers and authors of machine learning papers can collaborate and communicate to build on top of each other’s work with more ease and less frustration.

Read the article here.

If you’re interested in the latest in artificial intelligence research, read our reviews of the latest papers published online.