Scholar Maps: a journey from LiRA to the universe of human knowledge

There is an incredible new tool out there, called Scholar Maps. It's from the people behind Scholar Inbox, a trainable individualized recommender algorithm, that supplies you with daily digests of new arxiv papers.

Let's say we look up the paper from Nicholas Carlini et al. about Membership Inference Attacks From First Principles. We find it's the center of a topic called LiRA, short for Likelihood-Ratio Attack, which is the attack they invented in this paper! Nice!

We see the LiRA topic surrounded by other kinds of attacks, such as Label-Only Membership Inference Attacks. Since LiRA relies on the log-probabilities of the correct class, it needs more than the label. Seems correct.

The LiRA topic is part of the Machine Learning Security subfield. We see another topic a bit further away, such as the Model Inversion Attack.

Going out a bit further, we see the neighboring subfields: Privacy-Preserving Machine Learning and various seemingly synonymous wordings. Secure Computation seems pretty close, given it's a cryptography topic applicable to many use cases. Even from this distance, we see some highlight papers: Label-only membership inference attacks, PATE, PATE-GAN, GAN.

Leaving the subfields behind us, we see the larger picture of research fields: Machine Learning Security is part of the Data Privacy field. And it's close to other fields such as Software Engineering or Education.

It thinks we are part of the Academic domain of sociology. Fair enough, I guess, since privacy and security are societal issues that we aim to solve.

We are further away from our mathematics friends than it feels. But electrical engineering is quite close. Maybe it makes sense that so many people who studied EE switch to machine learning, like myself?

But yeah, looking at this it makes sense that if you take computer science and go further into the direction of philosophy and sociology, you might end up with data privacy. Cool!

And there you go. The universe of human knowledge (that exists on arxiv and is indexed by them, anyway).

Check out scholar-maps with some paper you care about:

Scholar Inbox

Scholar Inbox is a personal paper recommender which enables researchers to stay up-to-date with the most relevant progress in their field based on their personal research interests. Scholar Inbox is free of charge and daily indexes all of arXiv, bioRxiv, medRxiv and ChemRxiv as well as several open access proceedings in computer science. Register today and never miss a beat again!

Beyond just the fun of looking at it, this tool can help uncover blind spots. Unlike, say ConnectedPapers, it lets you explore 'similar' research without formal connections between the papers.