PhD Student at Mila Quebec AI Institute, and Université de Montréal.
Joined December 2021
- Tweets54
- Following391
- Followers571
- Likes355
Ryan D'Orazio retweeted
Applications for the JHU DSAI postdoctoral fellowship are open until January 26, 2026.
If you are interested in working with me in the intersection of Optimization and ML, please apply and reach out!
Join us in advancing data science and AI research! The Johns Hopkins Data Science and AI Institute Postdoctoral Fellowship Program is now accepting applications for the 2026–2027 academic year. Apply now! Deadline: Jan 23, 2026. Details and apply: apply.interfolio.com/179059
Ryan D'Orazio retweeted
We will some extra funds to hire world-class PhD students and PDFs; if you know such folks who are also interested in topics I care about (RL theory, and now LLM reasoning theory), please send them my way (or if you are one of those, please DM me).
I'll be at NeurIPS in San Diego, happy to chat and meetup! If you're interested in optimization or the Polyak stepsize come check out our poster with @bremen79!
Ryan D'Orazio retweeted
Our paper on Pseudo-Asynchronous Local SGD is accepted at TMLR!
Developed at @MSFTResearch, it introduces a semi-synchronous training strategy that pairs well with methods like DiLoCo.
Clean code & camera-ready coming soon. Thanks to my co-authors!
arxiv.org/abs/2504.18454
Ryan D'Orazio retweeted
I am at ICCOPT this week, at USC. Let's meet if you want to chat about optimization for machine learning and/or you are interested in working as a post-doc with me
On my way to ICCOPT I decided to give PEPit (github.com/PerformanceEstima…) a try, and I wish I had used it sooner. In just a few hours I was able to confirm our theoretical results in our recent paper and I was able to get intuition that originally took me months without using it. 1/N
These results also hint that our condition on the error is quite tight! As demonstrated by the bold red curve never crossing below the dotted red line for any stepsize. Looking forward to using this tool for other algorithms!
1
3
182
I’m also excited to be presenting this work (openreview.net/forum?id=4ZX2…) at ICCOPT at USC. Theory aside there are some applications that may interest ppl in RL, games, and performative prediction. Let me know if you are in the area and want to chat!
You can view a special case of our results using PEPit above. The main takeaway is that managing relative error in VIs is fundamentally more difficult than scalar min. For scalar min there is a step that can get you to converge for error < 1. But this is not true in VIs!
3
108
Polyak’s step is magical, it converges in iterate to a min of a cvx fun w/o Lipschitz assumptions. But, it is finicky, small changes might not converge. Our paper: arxiv.org/abs/2505.20219
provides new insight on the magic (stubbornness) via local (approx) curvature of a surrogate
Ryan D'Orazio retweeted
Happy to share that Compositional Risk Minimization has been accepted at #ICML2025
📌Extensive theoretical analysis along with a practical approach for extrapolating classifiers to novel compositions!
📜 arxiv.org/abs/2410.06303
This week I'll be at #ICLR25.
If you like fundamental optimization results, I'll be presenting our work on surrogate losses for non-convex-concave min-max problems and learning value functions in deep RL (VIs more generally).
Poster: #377 Thursday April 24 10am-12:30pm
I'll be at #NeurIPS24 until Sunday. If you're interested in solving variational inequality problems with deep learning (e.g. min-max and projected Bellman error), come and checkout our poster on surrogate losses at the opt ml workshop. arxiv.org/abs/2411.05228
I'll be at #NeurIPS24 until Sunday. If you're interested in solving variational inequality problems with deep learning (e.g. min-max and projected Bellman error), come and checkout our poster on surrogate losses at the opt ml workshop. arxiv.org/abs/2411.05228
Ryan D'Orazio retweeted
I'll be hiring 1-2 students for graduate positions in my group at Mila and the University of Montreal. Deadline to express your interest is December 1st!
mila.quebec/en/supervision-r…
Excited to share that I was selected as a 2023 @BorealisAI fellow and honored to be awarded alongside 9 other great researchers. This would not have been possible without support from my advisors, and many collaborators. I feel privileged to continue my research in optimization.
We are thrilled to introduce the 2023 @BorealisAI Fellowship recipients - exceptional researchers advancing #AI & #ML. Supported by Canada’s top institutions and leading researchers, they aim to make significant contributions to the field. Learn more: borealisai.com/news/the-bore…
5
35
4,627
Central to RL is dynamic programming – breaking down a sequential decision problem to a sequence of simpler single-step problems. Imperfect information in zero-sum games does not easily admit dynamic programming. Check out our work to see how we can fix it with regularization!
Imperfect information makes RL and search hard.
In our ICML paper, we give an equilibrium-preserving reduction from imperfect information adversarial games to fully observable games, opening the door to new RL methods and simpler search techniques. 1/N
arxiv.org/abs/2301.09159
7
992
Ryan D'Orazio retweeted
Major update to my Online Learning monograph
arxiv.org/abs/1912.13213
- Added Optimistic OMD
- Added chapter on saddle-point optimization using OCO algorithms
- Added chapter on universal portfolio
- More proofs
- Fixed bugs&typos
- Overall, I added 40+ pages
If you're at ICLR and interested in fast learning in games then make sure to reach out to @ssokota.
We provide some new results at the intersection of optimization, algorithmic game theory, and reinforcement learning. With contributions pushing the frontier in all three domains!
A longstanding disappointment in multi-agent research has been that naive self-play RL fails in adversarial games.
In our ICLR paper, we present a simple principled approach that enables self-play RL to succeed in such settings. 1/N
arxiv.org/abs/2206.05825
Ryan D'Orazio retweeted
Going to Neurips. Feel free to reach-out I'd you want to talk about postdoc or Ph.D. opportunities in my group at @Mila_Quebec exploring the interface between game theory, deep learning and optimization.