Alireza Mousavi

69 posts

Alireza Mousavi banner
Alireza Mousavi

Alireza Mousavi

@alirezamh_

CS PhD student @UofT and @VectorInst. Ex research intern @Apple ML Research. Interested in deep learning theory and generative modeling.

Toronto, Ontario Katılım Ekim 2024
273 Takip Edilen450 Takipçiler
Sabitlenmiş Tweet
Alireza Mousavi
Alireza Mousavi@alirezamh_·
Can RL with outcome rewards alone just efficiently explore outside the support of the base model and learn completely new capabilities? We study it in a theoretically tractable setup, and prove that outcome rewards are not enough, but process rewards can get you there! [1/3]
Alireza Mousavi tweet mediaAlireza Mousavi tweet mediaAlireza Mousavi tweet media
English
4
21
123
26.9K
Alireza Mousavi
Alireza Mousavi@alirezamh_·
I’m at ICML 🇰🇷 presenting a 🔦 spotlight 🔦 poster today. Check it out if you want to know how process rewards help RLVR go beyond the base model support when outcome rewards alone can’t. 📆 Thu. 5-6:45 pm. Hall A. Poster # 1307.
Alireza Mousavi tweet media
English
1
2
14
655
Alireza Mousavi retweetledi
Alberto Bietti
Alberto Bietti@albertobietti·
Attention and MLP matrices can store as much knowledge as # params, but in practice, finite samples/compute and long context noise can further limit this capacity. We characterize these trade-offs in a needle-in-a-haystack model. #ICLR2026 poster Thurs, led by @nurimertvural45
Alberto Bietti tweet mediaAlberto Bietti tweet mediaAlberto Bietti tweet media
English
1
25
153
12.8K
Alireza Mousavi retweetledi
Lily Goli
Lily Goli@lily_goli·
Excited to share @YaldaForoutan and @pekzta4's work on robust 3D reconstruction from 360 captures! Casual 360 video capture gives you the coverage you need for calibration and reconstruction; FullCircle makes it hassle-free! No more wrestling with stitching artifacts or floaters from the camera operator. Data and code are released, enjoy!
Andrea Tagliasacchi@taiyasaki

📢📢📢 Introducing "FullCircle: Effortless 3D Reconstruction from Casual 360° Captures" TL;DR: 10x faster casual capture with clean reconstructions Homepage: theialab.github.io/fullcircle Code: github.com/theialab/fullc… arXiv: arxiv.org/abs/2603.22572 Led by Yalda Foroutan & Ipek Oztas

English
0
7
28
5.8K
Alireza Mousavi
Alireza Mousavi@alirezamh_·
@thegautamkamath Wouldn’t the authors ultimately take a reputational hit once their paper is public (e.g. arxiv)?
English
0
0
2
514
Gautam Kamath
Gautam Kamath@thegautamkamath·
CS theory conferences used to be single blind, though recently (in the last 5-10 years) they moved to lightweight double blind. 1 benefit of single blind: authors can't submit trash without taking a reputational hit. It's increasingly clear that paper submission can't be "free."
English
8
0
129
13.9K
Alireza Mousavi
Alireza Mousavi@alirezamh_·
This is no longer an issue if you use process rewards. Number of reward queries will depend on a “token-level likelihood quantile” that is never exponentially small in sequence length. [3/3] 📄 arXiv: arxiv.org/abs/2603.06957
English
0
1
6
1K
Alireza Mousavi
Alireza Mousavi@alirezamh_·
We prove the number of reward queries in RLVR depends on the “likelihood quantile” function of the base model. Unfortunately, LQ becomes exponentially small if we want to go significantly below the base model error, then RL needs exponentially many reward queries. [2/3]
Alireza Mousavi tweet mediaAlireza Mousavi tweet media
English
1
1
5
1.3K
Alireza Mousavi
Alireza Mousavi@alirezamh_·
Can RL with outcome rewards alone just efficiently explore outside the support of the base model and learn completely new capabilities? We study it in a theoretically tractable setup, and prove that outcome rewards are not enough, but process rewards can get you there! [1/3]
Alireza Mousavi tweet mediaAlireza Mousavi tweet mediaAlireza Mousavi tweet media
English
4
21
123
26.9K
Alireza Mousavi retweetledi
Michael Kirchhof
Michael Kirchhof@mkirchhof_·
New paper 🥳 RL relies a lot on an agent’s capability to explore. Our strategy-guided exploration makes the agent find new solutions more efficiently. It learns faster, and in some environments its Pass@1 surpasses the base model’s Pass@128. 🧵1/6 📄 arxiv.org/abs/2603.02045
Michael Kirchhof tweet media
English
3
13
67
4.4K
Alireza Mousavi retweetledi
VAIBHAV SINGH
VAIBHAV SINGH@VAIBHAV22155287·
Masked Diffusion LMs (MDLMs) are the most exciting paradigm shift in AR generation because they can decode in parallel, infill, and self-correct. But they are bottlenecked by the transformer's quadratic attention, making throughput fall apart for long contexts. We offer a simple solution. Introducing DiffuMamba: first diffusion LM with a bidirectional Mamba backbone. Better quality. Up to 8.2x faster. 🧵1/N
English
4
27
162
15.5K
konglk1203
konglk1203@konglk1203·
@alirezamh_ Huge congrats! Incredible work. We happen to have a similar paper also accepted to ICLR26 arxiv.org/abs/2510.15038 Great minds think alike XD We plan to cite your work in our camera-ready. Hope we could chat together one day!
English
1
0
2
229
Alireza Mousavi
Alireza Mousavi@alirezamh_·
Now accepted at ICLR 2026. Check it out if you’re a diffusion/flow/OT enjoyer.
Alireza Mousavi@alirezamh_

Super excited to share what @stephenz_y and I’ve been up to during our internship at🍎: Using optimal transport makes flows straighter and generation faster in flow matching, but small batch OT is biased and large batch OT is slow. What to do? Use semidiscrete OT! 🧵

English
3
8
129
22.4K
Mathieu
Mathieu@miniapeur·
What are some of the most interesting recent papers in the theory of deep learning (about LLMs, foundation models, diffusion models, etc.)?
English
4
2
25
3.8K
Alireza Mousavi retweetledi
stephen zhang
stephen zhang@stephenz_y·
- Sat Dec 6 @ SPIGM workshop, On Fitting Flow Models with Large Sinkhorn Couplings, work with @alirezamh_ , Michal Klein, Marco Cuturi 4/4
stephen zhang tweet media
English
0
1
4
454
Alireza Mousavi retweetledi
Michael Kirchhof
Michael Kirchhof@mkirchhof_·
Our research team is hiring PhD interns 🍏 Spend your next summer in Paris and explore the next frontiers of LLMs for uncertainty quantification, calibration, RL and post-training, and Bayesian experimental design. Details & Application ➡️ jobs.apple.com/en-my/details/…
English
8
56
337
55.3K
Alireza Mousavi
Alireza Mousavi@alirezamh_·
For my usual ML theory math questions Gemini 3 doesn't feel different from GPT 5.1, both can answer small very well-defined questions and fail otherwise. Maybe the 2x improvement on ARC-AGI-2 isn't enough to replace mathematicians yet 🤔
English
0
0
7
557

Keşfet

@nurimertvural45 @YaldaForoutan @pekzta4 @thegautamkamath @konglk1203 @mufan_li @roydanroy @miniapeur