Improving Deep Learning With a Little Help From Physics | Quanta Magazine Rose Yu has a plan for how to make AI better, faster and smarter — and it’s already yielding results.

Rose Yu (@yuqirose.bsky.social) has a plan for how to make AI better, faster and smarter — and it’s already yielding results. Steve Nadis reports: www.quantamagazine.org/improving-de...

Thank you David! It would be great to meet on campus sometime ;)

Thank you! Great to know that. I’ll take a look. About creativity I have been thinking about this. On one hand, generative models aim to max the training set likelihood, on the other hand, we want the samples to be out of distribution…any good pointers to resolve this dilemma?

My #NeurIPS204 workshop keynote: "Analytic theory of creativity of convolutional diffusion models" now available: neurips.cc/virtual/2024...
It reports on work with @MasonKamb in this paper: arxiv.org/abs/2412.20292

Beautiful talk!!! It seems that this analysis can also extend to group convolutions? Another question is how to quantify creativity?

Hello there! We work on machine learning algorithms for epidemic forecasting and prevention!

Speaking of healthcare. Let's make @bsky.app the platform that fights misinformation. Some great follows are @drsusanoliver.bsky.social @jsm2334.bsky.social @therealtruther.bsky.social

Let's get them more followers and engagement than other platforms and let health science accounts find a home

That is a good point! The main responsibility of AC is to facilitate discussions among reviewers. But without much at stake, reviewers don’t have a strong incentive to reach consensus.

That's an interesting perspective. What is the ideal effect that we want from such reviewer's discussion? Don't we want independent opinions from each reviewer?

Hello world, Bluesky! PhD hiring + Faculty hiring. It's going to be a busy season of 2025!