New preprint!
This is a hardcore technical paper on Thompson sampling - as a strategy for the so-called online learning game.
I think it's one of the most long-term important things I have ever worked on due to what it makes possible.
But that needs explaining - thread below!
I’m excited to announce that in July 2025 I will be joining @UWaterloo as an Assistant Professor in the Department of Statistics and Actuarial Science! Until then, I will continue at Princeton as a DataX Postdoc Fellow, working with Boris Hanin.
I have many exciting projects lined up to study the scaling limits of neural networks and their training dynamics, with some important practical implications on the horizon. Stay tuned!
@roydanroy@ArnaudDoucet1@wellingmax I think the analogy needed to make ``optimal scaling'' work is that each of your papers is an infinite product of block-iid sections.
Congratulations to Dr @HaghifamMahdi on a successful thesis defense at U of T!
🌟🔥🎉👏🔥🌟🎉
Next stop: Distinguished Postdoctoral Fellow at the Khoury College of Computer Sciences at @NorthEastern.
3.5 years after we started this project while visiting @the_IAS, it’s been accepted to the Annals of Statistics.
I won’t be at #JSM2023 but make sure to stop by Jeff’s session to hear about our work!
Super excited that this work has been accepted for publication in AoS.
For those at #JSM2023, I'll be speaking about it in the session on Advances in Statistical Machine Learning, scheduled for Sunday 8/6/2023 2:00:00 PM.
I'm speaking first so don't be late 😉.
I'm _incredibly_ excited about this paper.
Deep insights for prompting, and for how models represent latent knowledge about the world. See the paper and check out @wzihao12's thread!
Some extra (more speculative) nuggets below:
Prompt engineering is a dark art. To understand limitations, we formalize "what the user intended" through latent concepts. Then we show direct model controls by algebraic operations on suitably chosen representations!https://t.co/kK4I38Jf76 w @ybnbxb@jeffNegrea@victorveitch
@roydanroy it expects labels to be all caps sometimes, and it seems safer to wrap \crefs in {} within titles. doing that and defining the label twice, once with all caps and works for me... dm me a mwe if not