Tam Le

1/ Where do Probabilistic Models, Sampling, Deep Learning, and Natural Sciences meet? 🤔 The workshop we’re organizing at #NeurIPS2025! 📢 FPI@NeurIPS 2025: Frontiers in Probabilistic Inference – Learning meets Sampling Learn more and submit → fpiworkshop.org Organizers: @tara_aksa @msalbergo @bose_joey @marylougab @theh2o64 @guanhorng_liu Grant Rotskoff @EvaSmorodina @AlexanderTong7

🎉 Happy to share that our work "Flowing Datasets with Wasserstein over Wasserstein Gradient Flows" was accepted at #ICML2025 as an oral! This is a joint work with the amazing Christophe Vauthier and @Korba_Anna! Link: openreview.net/forum?id=I1OHP…

[3] 17 Jul 11 a.m. PDT — 1:30 p.m. PDT (East Exhibition Hall A-B # E-1500) Scalable Sobolev IPM for Probability Measures on a Graph TL*, Truyen Nguyen*, Hideitsu Hino, Kenji Fukumizu

We will present 3 posters at ICML’25: [1] Tree-Sliced Wasserstein Distance: A Geometric Perspective [2] Tree-Sliced Wasserstein Distance with Nonlinear Projection [3] Scalable Sobolev IPM for Probability Measures on a Graph Please come to our posters for more details. Thanks!

[2] 16 Jul 4:30 p.m. PDT — 7 p.m. PDT (East Exhibition Hall A-B # E-3412) Tree-Sliced Wasserstein Distance with Nonlinear Projection Thanh Tran*, Hoang V. Tran*, Thanh Chu, Trang Pham, Laurent El Ghaoui⊛, TL⊛, Tan M. Nguyen⊛.

[1] 15 Jul 4:30 p.m. PDT — 7 p.m. PDT (West Exhibition Hall B2-B3 # W-1004) Tree-Sliced Wasserstein Distance: A Geometric Perspective Hoang V. Tran*, Trang Pham*, Tho Tran, Khoi Nguyen, Thanh Chu, TL⊛, Tan M. Nguyen⊛.

🎖The 2025 Kyoto Prize Announcement Live Streamed at Around 3:00 p.m. Today! youtube.com/live/5wiyHbTH6… #the2025KyotoPrize

You can find all our lectures on YouTube (thanks to @StanfordOnline): youtube.com/playlist?list=… and the assignments on the course website so you can do it yourself at home: stanford-cs336.github.io/spring2025/

Great videos! I learned so much. The assignments are too hard for me.

BREAKING NEWS Congratulations to Professor Shun-ichi Amari! 2025 Kyoto Prize Laureates kyotoprize.org/en/laureates/s…

📢📢📢 𝐓𝐫𝐢𝐚𝐧𝐠𝐥𝐞 𝐒𝐩𝐥𝐚𝐭𝐭𝐢𝐧𝐠 𝐟𝐨𝐫 𝐑𝐞𝐚𝐥-𝐓𝐢𝐦𝐞 𝐑𝐚𝐝𝐢𝐚𝐧𝐜𝐞 𝐅𝐢𝐞𝐥𝐝 𝐑𝐞𝐧𝐝𝐞𝐫𝐢𝐧𝐠 arxiv.org/abs/2505.19175 trianglesplatting.github.io TL;DR: use triangles rather than Gaussians, and everything works just as well. Code (soon) github.com/trianglesplatt…

If you reviewed for #ICML, please make sure to read other reviewers' comments too and reflect on whether you may have missed something. The paper will need to have a single decision; the point of rebuttal is not just about addressing each reviewer's concerns individually.

'High-Dimensional Probability' - Vershynin (2024, PDF): math.uci.edu/~rvershyn/pape… Full course on high-dimensional probability with 41 video lectures & 13 problem sets provided: math.uci.edu/~rvershyn/teac…

Transformers Learn to Implement Multi-step Gradient Descent with Chain of Thought

Oooooh. A rebuttal acknowledgment! I feel so special! Someone clicked a button!

A free book: Learning Theory from First Principles by @BachFrancis It covers a bunch of key topics from machine learning (ML) theory and practice, such as: - Math basics - Supervised learning - Generalization, overfitting & adaptivity - Tools to design learning algorithms - Optimization in ML - Local, Kernel and sparse methods - Neural networks - Ensembles - Online learning - Overparameterized models and more! The book also includes simple experiments (in MATLAB and Python), exercises, and references to more advanced material Read it here: di.ens.fr/~fbach/ltfp_bo…

The Norwegian Academy of Science and Letters has decided to award the Abel Prize 2025 to professor Masaki Kashiwara. Kashiwara-san is the first Japanese mathematician to receive the Abel Prize. #Abelprize #Abelprize2025 #mathematics #science

Traditional autoregressive models with Chain-of-Thought reasoning are inefficient and prone to error accumulation. This paper introduces Diffusion of Thought (DoT). DoT leverages diffusion models to perform Chain-of-Thought reasoning. 📌 Diffusion models offer parallel reasoning steps unlike autoregressive token-by-token generation. 📌 DoT achieves flexible computation cost by adjusting diffusion timesteps for reasoning tasks. 📌 Scheduled sampling in DoT enhances self-correction, addressing error accumulation in reasoning. ---------- Methods Explored in this Paper 🔧: → DoT uses a diffusion model. It generates reasoning steps in parallel across diffusion timesteps. → This contrasts with autoregressive models' token-by-token approach. → DoT employs classifier-free guidance for conditioning on queries. This ensures better control over tokens. → Training includes scheduled sampling. It improves self-correction by exposing and correcting errors from prior steps. → Multi-pass DoT generates thoughts sequentially. This introduces causal bias. → A conditional Ordinary Differential Equations solver accelerates inference. This speeds up continuous diffusion models. → DoT demonstrates flexibility in balancing computation and reasoning performance. → Self-consistency decoding further enhances DoT's accuracy. ---------------------------- Paper - arxiv. org/abs/2402.07754 Paper Title: "Diffusion of Thoughts: Chain-of-Thought Reasoning in Diffusion Language Models"

Following a thorough investigation, the Program Chairs (PCs) decided to desk-reject 19 papers authored by confirmed highly irresponsible reviewers, which would have been accepted otherwise, in accordance with the previously communicated CVPR 2025 policies. 2/2

#CVPR2025 Area Chairs (ACs) identified a number of highly irresponsible reviewers, those who either abandoned the review process entirely or submitted egregiously low-quality reviews, including some generated by large language models (LLMs). 1/2