Associate prof. at the University of Montréal and Mila. Research scientist Google DeepMind. Previously Stanford; UT Austin.
Montréal / Athens
Joined June 2013
- Tweets2K
- Following873
- Followers4.2K
- Likes4.7K
Ioannis Mitliagkas (Γιάννης Μητλιάγκας) retweeted
(1/n) 🎉 Big news for Archimedes at @NeurIPSConf 2026!
🔗 Explore more publications by Archimedes members: archimedesai.gr/en/publicati….
#NeurIPS2026 #ArchimedesUnit #AthenaResearchCenter #ArtificialIntelligence #AI #MachineLearning #Research
Ioannis Mitliagkas (Γιάννης Μητλιάγκας) retweeted
Thrilled to announce our $40M fundraise! We would like to thank our investors (Wing, Mayfield, 8VC, The House Fund) and angels (Jeff Dean, Dheeraj Pandey, Tristan Handy), Tasso Argyros, and several others from Anthropic, OpenAI, Meta, and our customers in frontier labs and enterprises that we work with every day. Bespoke Labs will make AI agents reliable and make it easier to optimize their performance.
We’re thrilled to announce a $40M investment that will fuel our mission to make AI agents reliable.
For the past two years, we've been heads-down doing world-class data curation research and shipping best-in-class reinforcement learning environments for training and optimizing AI agents. This funding lets us go a lot deeper on both.
Thank you to our investors @Wing_VC, @MayfieldFund, @8vc, @thehousefund and our angels such as Jeff Dean, Dheeraj Pandey, Tristan Handy, and several others from Anthropic, OpenAI, Meta. And thanks to the frontier labs and enterprises we work with every day, for sharing our vision for a future where agents can run autonomously for weeks and months at a time.
(more below)
Ioannis Mitliagkas (Γιάννης Μητλιάγκας) retweeted
#ICLR2026 🇧🇷
If you are interested in finding out why weights of neural networks discover the hidden signal structure, come see us at Pavilion 3 #815.
This is joint work with @yorgos_pantis, @bouzoukipunks and @ChristosTzamos.
Ioannis Mitliagkas (Γιάννης Μητλιάγκας) retweeted
Earlier this month I defended my PhD at Mila / Université de Montréal:
"Toward Efficient and Scalable Optimization: Theoretical Insights and Practical Challenges."
Grateful to my advisor Ioannis Mitliagkas ( @bouzoukipunks ), and to my committee:
Simon Lacoste-Julien ( @SimonLacosteJ ) , Gauthier Gidel ( @gauthier_gidel ), and Aurélien Lucchi ( @AurelienLucchi ).
This summer I'm continuing in that direction at NVIDIA (Redmond), working on efficient LLM training and training infrastructure.
Ioannis Mitliagkas (Γιάννης Μητλιάγκας) retweeted
What's the Critical Batch Size for optimizers like Muon & Signum? Their geometry isn't Euclidean, so the usual gradient noise scale (GNS) doesn't fit.
Our ICML 2026 paper (Mila × Meta): measure GNS in the optimizer's dual norm, then grow the batch size during training — reaching the same loss in fewer steps.
Details in @paramsraman's thread 👇 Come by the poster:
📍 Wed Jul 8, 2:30–4:15 PM · Hall A #3704
openreview.net/forum?id=XMSa…
1/11 How should the gradient noise scale (GNS) be defined for non-Euclidean optimizers, such as SignSGD and Muon? Key insights from our recent work on extending GNS to stochastic sign and spectral descent; see arxiv.org/abs/2602.03001.
Ioannis Mitliagkas (Γιάννης Μητλιάγκας) retweeted
How does public data affect unlearning guarantees?
Come find out RIGHT NOW in HALL A #3110 (10:30 AM)
"Unlearning with Asymmetric Sources: Improved Unlearning-Utility Trade-off with Public Data,"
Can't come? Enjoy this ICML'26 Seoul🇰🇷 tweet 🧵...
Ioannis Mitliagkas (Γιάννης Μητλιάγκας) retweeted
Presenting our work at ICML this week! Come and say hi!
On the Sample Efficiency of Inverse Dynamics Models for Semi-Supervised Imitation Learning
Poster: Wed, Jul 8, 10:30 AM KST, HALL A # 508
Action labels for BC are not cheap. Semi-supervised methods use Inverse Dynamics Models (IDMs) to fix this, but why does IDM learning actually outperform BC in this context?
Our #ICML2026 paper answers this! 🇰🇷
📜 Paper: arxiv.org/abs/2602.02762
🧵
Ioannis Mitliagkas (Γιάννης Μητλιάγκας) retweeted
Excited to share that our group has 3 papers at #ICLR2026!
2 main conference papers + 1 oral at GRaM Workshop on Geometry-grounded Representation Learning
Topics: model merging, fairness in task arithmetic, and geometry of Muon vs SignSGD 🧵👇
(1/8)
Ioannis Mitliagkas (Γιάννης Μητλιάγκας) retweeted
[1/8] In-context learning and prior-fitted networks are gaining traction for causal effect estimation in tabular datasets.
But can they also generate novel observational & interventional data from causal models?
📌Introducing Cond-FiP: In-context learning of causal mechanisms via conditional fixed-point iterations.
📜 arxiv.org/abs/2410.06128
🔹Accepted at TMLR and to be presented at ICLR 26 in the J2C track!
Another Turing award for our department at @UMontreal this time for Gilles Brassard!
With colleagues like Gilles and @Yoshua_Bengio (just to name a few) I'm afraid that my imposter syndrome will never dissipate.
Congratulations to Charles H. Bennett (@IBMResearch) and Gilles Brassard ( @UMontreal) on receiving the 2025 ACM A.M. Turing Award!
🔗: awards.acm.org/turing
Ioannis Mitliagkas (Γιάννης Μητλιάγκας) retweeted
1/4 Excited to share that our paper has been accepted at ICLR 2026! 🎉
Preprint available here: arxiv.org/abs/2510.19382
Ioannis Mitliagkas (Γιάννης Μητλιάγκας) retweeted
If you're a senior researcher (i.e., associate professor or more senior), and you'd consider moving to the University of Toronto to take up a Canada Impact+ Research Chair (1 million/year in funding), email me. Discretion guaranteed.
++Please share to help me spread the word!++
Canada is investing over $130 million to recruit top researchers from abroad, with up to 1 million/year in research funding.🚀
📢Also: 600 doctoral scholarships and 400 postdoctoral research awards!
Join us at UToronto: jobs.utoronto.ca/job/Toronto…
@UofT @VectorInst @TorontoSRI
Ioannis Mitliagkas (Γιάννης Μητλιάγκας) retweeted
[1/9] While pretraining data might be hitting a wall, novel methods for modeling it are just getting started!
We introduce future summary prediction (FSP), where the model predicts future sequence embeddings to reduce teacher forcing & shortcut learning.
📌Predict a learned embedding of the future sequence, not the tokens themselves
Ioannis Mitliagkas (Γιάννης Μητλιάγκας) retweeted
🚨 New paper!
“Understanding Adam Requires Better Rotation-Dependent Assumptions.”
Come check out our poster at @NeurIPSConf, or DM me if you would like to chat!
📅 Wednesday, December 3
🕐 4:30 PM PST
📍Exhibit Hall C,D,E #908
Ioannis Mitliagkas (Γιάννης Μητλιάγκας) retweeted
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!
Ioannis Mitliagkas (Γιάννης Μητλιάγκας) retweeted
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
Ioannis Mitliagkas (Γιάννης Μητλιάγκας) retweeted
Presenting CRM at #ICML2025
📌 Wednesday, 16th July, 11 am
📍East Exhibition Hall A-B (E-2101)
Lets chat about distribution shifts! Been deep into causality & invariance based perspectives, and recently exploring robust LLM pretraining architectures.
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
Ioannis Mitliagkas (Γιάννης Μητλιάγκας) retweeted
I'm delighted to share that our paper has been accepted
by #TMLR!
We empirically observed signs of scaling laws regarding how the choice of pre-trained models affects OOD test errors and Expected Calibration Error on downstream tasks.
An Empirical Study of Pre-trained Model Selection for Out-of-Distribution Generalization and Cali...
Hiroki Naganuma, Ryuichiro Hataya, Kotaro Yoshida, Ioannis Mitliagkas.
Action editor: Mingsheng Long.
openreview.net/forum?id=tYjo…
#accuracy #trained #deep
Ioannis Mitliagkas (Γιάννης Μητλιάγκας) retweeted
My thesis is now online!
umontreal.scholaris.ca/items…
This is more than just a list of publications. I invested a lot of time and passion writing this thesis in hope that it will make for an interesting read.
Here's a summary of what you'll find in it.
