Ryan Adams
@ryan_p_adams
Followers
35K
Following
4K
Media
18
Statuses
1K
Machine Learning Researcher, CS Professor (@PrincetonCS), Dad, Woodworker
Joined January 2015
Missing ICML due to visa :'(, but looking forward to share our ICML paper ( https://t.co/gXWBQNkgRs) as a poster at #BayesComp, Singapore! Work on symmetrising neural nets for schrodinger equation in crystals, with the amazing Zhan Ni, Elif Ertekin, Peter Orbanz and @ryan_p_adams
arxiv.org
Incorporating group symmetries into neural networks has been a cornerstone of success in many AI-for-science applications. Diagonal groups of isometries, which describe the invariance under a...
0
6
47
Congrats to Kai Li on being named a member of the American Academy of Arts & Sciences! 🎉 Li joined @Princeton in 1986 and has made important contributions to several research areas in computer science. https://t.co/gYVjWnHX8J
1
16
108
Not sure why the gutting of American science funding isn’t a bigger story. No one voted for it, it reduces American innovation and economic competitiveness in the near-term and long-term, and it isn’t even being done efficiently, if that were in fact the goal.
250
951
5K
No.
At first glance, bar charts might seem like a simple visualization type. But with a little creativity, they can be enhanced in countless ways to reveal deeper insights and make your data shine. From summarizing counts to comparing proportions, bar charts are a powerful and
13
25
273
You need to understand both General Relativity and Hubble expansion to correctly engineer Global Positioning System, which is a central part of modern civilian and military infrastructure. Here's why. 1/
33
162
1K
Since adaptive tokenisation is trendy these days, this paper from a decade ago (an absolute eternity in DL ⌛️) is worth revisiting! By @oren_rippel, Michael Gelbart and @ryan_p_adams
https://t.co/bNUYlBpsou
arxiv.org
In this paper, we study ordered representations of data in which different dimensions have different degrees of importance. To learn these representations we introduce nested dropout, a procedure...
reading the nested dropout paper, which is excellent and presents a way to learn an ordered latent representation (the authors later founded a ML compression company that was acquired by Apple)
3
11
121
#PrincetonU has launched AI for Accelerating Invention. Led by professors @MengdiWang10 and @ryan_p_adams, the AI^2 initiative aims to achieve faster breakthroughs across engineering disciplines: https://t.co/fkubSlpKWm
0
5
33
We are looking for AI Postdoctoral Fellows! Be a part of AI for Accelerating Invention, a new research initiative out of the Princeton AI Lab. You can learn about AI for Accelerating Invention here: https://t.co/BYUplNNyKq and apply for the postdoc here: https://t.co/iDM0IzE1hM.
5
20
77
BREAKING NEWS The Royal Swedish Academy of Sciences has decided to award the 2024 #NobelPrize in Physics to John J. Hopfield and Geoffrey E. Hinton “for foundational discoveries and inventions that enable machine learning with artificial neural networks.”
1K
13K
33K
@Wikipedia @GatsbyUCL I don't think it's a controversial page:
2
1
7
I find it strange that there's no @Wikipedia page for @GatsbyUCL, despite its huge influence on ML/AI/CompNeuro. I've tried twice to create a basic one and gotten rejected both times for "lack of reliable sources".
4
5
56
Our new paper in @NatMachIntell tells a story about how, and why, ML methods for solving PDEs do not work as well as advertised. We find that two reproducibility issues are widespread. As a result, we conclude that ML-for-PDE solving has reached overly optimistic conclusions.
7
107
491
Our recent work on developing a "physics aware" (or convex hull aware) active search method to more efficiently discover stable materials is now published in Materials Horizons! Link to PDF: https://t.co/GuLgAiLLCU
0
9
37
Excited to share CryoBench🧊🪑 our dataset and benchmarking effort for heterogeneous cryo-EM reconstruction! Led by @MinkyuJeon19791, who is an absolute machine, and super fun collab with @PilarCossio2 @sonyahans groups @FlatironInst A few thoughts on our benchmark design 👇
🚀 Excited to share our preprint: CryoBench 🧊🪑: Diverse & Challenging Datasets for the Heterogeneity Problem in Cryo-EM! 🌐 https://t.co/qssgR7hlWd 🙌 Huge thanks to my amazing co-authors, collaborators, & advisor @zhongingalong! 🎉 #cryoEM #MachineLearning 🧵 Details below!
1
32
119
We are about one election cycle away from prediction markets being manipulated by state actors in the way social media was in 2016.
2
5
20
Excited to present Generative Marginalization Models (MAMs) at #ICML2024! MAM trains a neural network for fast estimation of arbitrary marginals in discrete data, bypassing the need to eval. a sequence of conditional probs in any-order-ARMs. w/ Peter Ramadge, @ryan_p_adams 1/7
8
13
72
This setup then inherits guarantees from classic results on optimization under partial asynchrony. It also turns out to be a useful graph neural network architecture even outside the asynchronous setting. https://t.co/f1YvohmoRV I'm really excited about this work! 4/4
arxiv.org
Message passing graph neural networks (GNNs) would appear to be powerful tools to learn distributed algorithms via gradient descent, but generate catastrophically incorrect predictions when nodes...
0
9
60
To fix this, Olga Solodova has led a project on “Graph Neural Networks Gone Hogwild” (h/t @beenwrekt), studying this effect and proposing a solution. This work introduces an architecture in which the GNN minimizes a separable convex function. 3/4
2
2
29
The asynchrony effectively alters the architecture at inference time. 2/4
1
1
29