GroundAI revolutionizes scholarly peer review for AI research
CDS alum Andy Gan and NYU Wagner student Taha Raslan discussed their recent contribution to Machine Learning with Sabrina de Silva. After their own firsthand experience trying to engage with the academic community, Gan and Raslan joined forces to pioneer a peer review platform called GroundAI. Here, they discuss their work and what changes we can hope for with GroundAI’s first steps into the world of Machine Learning research.
What is GroundAI?
GroundAI is a web platform for scholars to keep up-to-date with the latest Machine Learning papers and raise discussions about them.
Can you discuss the inspiration behind GroundAI?
During our Deep Learning Course with Professor Yann LeCun, we could not find a place to share our final project paper, receive specialized feedback for further development, read other students’ work, and raise discussions about other academic papers.
We wanted an environment suited to all the different voices and perspectives. We needed a community that cherishes collaborations among like-minded people with similar passions and distinct ideas regarding AI. Therefore, we came up with Ground AI.
Could you give us the “tagline” of GroundAI? The central tenets that underpin its mission?
A Community Peer Review Platform.
Our mission is to increase AI scholarly communication. We want to help the community understand how a given paper’s experiments, data, and methods work, and promote collaboration among peers to improve existing research.
You mention that “preprints discussions usually happen on Twitter and Facebook.” Do you think that using GroundAI could limit research’s visibility and therefore inhibit relevant contributions from other domains’ experts?
GroundAI does not presently offer visibility in the same way that Twitter or Facebook might. The discussions and feedback from GroundAI, however, are of higher quality for two key reasons. Firstly, the platform is specialized because it is dedicated to academia. GroundAI enforces a content policy, which explicitly defines what constitutes a valuable contribution. All contributions are welcome, no matter what domain the contributor is from (AI, statistics, healthcare, physics, etc.), so long as they are relevant to advancing AI research. Secondly, discussions are stored directly alongside the preprint, and are openly accessible to all. This layout increases transparency and, by extension, the public’s overall understanding.
How would you like to see people use GroundAI?
We understand that the peer review process is a huge burden in the current ML publishing climate. The pool of submissions is rapidly growing, but the pool of experienced reviewers is comparatively stagnant. This has resulted in a lack of qualified reviewers.
We hope that the Machine Learning community embraces GroundAI as a new community peer review platform for faster, more efficient interaction with high quality feedback earlier in the research process to improve the work and expedite subsequent publications. We want this dialogue to become part of the scholarly record. High quality reviews help scientists to improve their work and expedite subsequent publication.
We also hope to see researchers use GroundAI to share data, code, posters, and supplementary materials to advance reproducibility of AI research. This will allow independent research teams to reproduce results via the same AI method, data, and tools made by the original research team.
What makes the preprint dialogue and revision process worthy of addition to the scholarly record?
We believe scientists should receive “credit” for peer review. It may take a cultural change for them to devote attention to preprints. Researchers’ competence should not only be judged by how many papers they have published, or how significant their contributions are, but also by their ability to critique others’ works, discover their flaws, and point out potential improvements on methodologies and experiments.
Are you currently working on anything new for the platform that we can look forward to?
We’re constantly improving the platform and pushing new features with the hope of developing new norms and best practices.
How have you seen GroundAI contribute to the public discourse?
GroundAI is very new and not a lot of people know about it yet. But, our current users love it and use it on a daily basis to keep up-to-date with the latest ML papers. Users can also follow their favorite authors to receive notifications when they publish new papers. We believe that the interface provides an unparalleled reading experience.
Users add ML papers to their library, provide feedback to authors, and search papers that use a specific network architecture like GAN, dataset like SQuAD, or cite authors like Ian Goodfellow (since Arxiv search is only limited to titles and abstracts).
By: Sabrina de Silva
CDS does not necessarily endorse or recommend the companies featured in this blog.