Aditya Arun

The Principles of Diffusion Models This monograph presents the core principles that have guided the development of diffusion models, tracing their origins and showing how diverse formulations arise from shared mathematical ideas. Diffu...

"The Principles of Diffusion Models" by Chieh-Hsin Lai, Yang Song, Dongjun Kim, Yuki Mitsufuji, Stefano Ermon. arxiv.org/abs/2510.21890
It might not be the easiest intro to diffusion models, but this monograph is an amazing deep dive into the math behind them and all the nuances

Video models are zero-shot learners and reasoners Fascinating new paper from Google DeepMind which makes a very convincing case that their Veo 3 model - and generative video models in general - serve a similar role in …

Made some notes on the new DeepMind paper "Video models are zero-shot learners and reasoners" - it makes a convincing case that generative video models are to vision problems what LLMs were to NLP problems: single models that can solve a wide array of challenges simonwillison.net/2025/Sep/27/...

As AI agents face increasingly long and complex tasks, decomposing them into subtasks becomes increasingly appealing.

But how do we discover such temporal structure?

Hierarchical RL provides a natural formalism-yet many questions remain open.

Here's our overview of the field🧵

What is the probability of an image? What do the highest and lowest probability images look like? Do natural images lie on a low-dimensional manifold?
In a new preprint with Zahra Kadkhodaie and @eerosim.bsky.social, we develop a novel energy-based model in order to answer these questions: 🧵

pretty sweet:

chewisinho.github.io/opt_notes_fi...

Optimal Transport for Machine Learners Optimal Transport is a foundational mathematical theory that connects optimization, partial differential equations, and probability. It offers a powerful framework for comparing probability distributi...

I have cleaned a bit my lecture notes on Optimal Transport for Machine Learners arxiv.org/abs/2505.06589

Happy to be recognized among outstanding reviewers 🙂

Enjoying reading this. Clarifies some nice connections between scoring rules, probabilistic divergences, convex analysis, and so on. Should read it even more closely, to be honest!

Nice tutorial arxiv.org/abs/2503.21673

"Inductive Moment Matching" by Linqi Zhou et al. I like the use of multiple particles to apply a loss similar to consistency models, but on distributions. Training is stable and gives high-quality generated images in very few sampling steps

📄 arxiv.org/abs/2503.07565
🌍 lumalabs.ai/news/inducti...

A Bear Case: My Predictions Regarding AI Progress — LessWrong This isn't really a "timeline", as such – I don't know the timings – but this is my current, fairly optimistic take on where we're heading. …

Smashing the endorse button as fast as I can
www.lesswrong.com/posts/oKAFFv...

Yet another short paper on randomness predictors (of which conformal predictors are a subclass) by Vovk, focusing on an inductive variant. arxiv.org/abs/2503.02803

I already advertised for this document when I posted it on arXiv, and later when it was published.

This week, with the agreement of the publisher, I uploaded the published version on arXiv.

Less typos, more references and additional sections including PAC-Bayes Bernstein.

arxiv.org/abs/2110.11216

1/n🚀If you’re working on generative image modeling, check out our latest work! We introduce EQ-VAE, a simple yet powerful regularization approach that makes latent representations equivariant to spatial transformations, leading to smoother latents and better generative models.👇

Slides for a general introduction to the use of Optimal Transport methods in learning, with an emphasis on diffusion models, flow matching, training 2 layers neural networks and deep transformers. speakerdeck.com/gpeyre/optim...

I don’t get the DeepSeek freak-out. Chinese orgs have been making models of all sorts that were on par with those of US orgs for a while now.

Convergence of Statistical Estimators via Mutual Information Bounds Recent advances in statistical learning theory have revealed profound connections between mutual information (MI) bounds, PAC-Bayesian theory, and Bayesian nonparametrics. This work introduces a novel...

We uploaded yesterday on arXiv a paper on a variant of the "Mutual Information bound" taylored to analyze statistical estimators (MLE, Bayes and variational Bayes, etc).

I assume I should advertise for it after the holidays, but in case you are still online today:
arxiv.org/abs/2412.18539

100%

which reminded me of @ardemp.bsky.social rule #1 on how to science: Don’t be too busy

Being too busy (with noise) = less time to read papers, less time to think and to connect the dots, less time for creative work!

Reinforcement Learning: An Overview

This manuscript gives a big-picture, up-to-date overview of the field of (deep) reinforcement learning and sequential decision making, covering value-based RL, policy-gradient methods, model-based methods, and various other topics.

arxiv.org/abs/2412.05265

I've been around the block a few times. When deep learning first became hot, many older colleagues bemoaned it as just tinkering + chain rule, and not intellectually satisfying. Then came SSL, equivariance, VAEs, GANs, neural ODEs, transformers, diffusion, etc. The richness was staggering.

🧵👇

Detection, Pose Estimation and Segmentation for Multiple Bodies: Closing the Virtuous Circle

Miroslav Purkrabek, Jiri Matas

tl;dr: detect bbox -> mask -> estimate human pose -> mask them and repeat. SAM-enabled method :)
arxiv.org/abs/2412.01562

Samples y | x from Treeffuser vs. true densities, for multiple values of x under three different scenarios. Treeffuser captures arbitrarily complex conditional distributions that vary with x.

I am very excited to share our new Neurips 2024 paper + package, Treeffuser! 🌳 We combine gradient-boosted trees with diffusion models for fast, flexible probabilistic predictions and well-calibrated uncertainty.

paper: arxiv.org/abs/2406.07658
repo: github.com/blei-lab/tre...

🧵(1/8)

Proud to announce our NeurIPS spotlight, which was in the works for over a year now :) We dig into why decomposing aleatoric and epistemic uncertainty is hard, and what this means for the future of uncertainty quantification.

📖 arxiv.org/abs/2402.19460 🧵1/10

Optimal transport computes an interpolation between two distributions using an optimal coupling. Flow matching, on the other hand, uses a simpler “independent” coupling, which is the product of the marginals.

Our big_vision codebase is really good! And it's *the* reference for ViT, SigLIP, PaliGemma, JetFormer, ... including fine-tuning them.

However, it's criminally undocumented. I tried using it outside Google to fine-tune PaliGemma and SigLIP on GPUs, and wrote a tutorial: lb.eyer.be/a/bv_tuto.html

Anne Gagneux, Ségolène Martin, @quentinbertrand.bsky.social Remi Emonet and I wrote a tutorial blog post on flow matching: dl.heeere.com/conditional-... with lots of illustrations and intuition!

We got this idea after their cool work on improving Plug and Play with FM: arxiv.org/abs/2410.02423

Updated: 6 benchmarks testing spatial and agent reasoning of LLM/VLMs
arxiv.org/abs/2410.06468 does spatial cognition
arxiv.org/abs/2307.06281 MMBench
arxiv.org/abs/2411.13543 BALROG
arxiv.org/abs/2410.07765 GameTraversalBenchmark
3dsrbench.github.io 3DSRBenchmark
open-eqa.github.io Open-EQA

Yeah, looks like different image resolutions will give you entirely different depth maps.

Theoretical Foundations of Conformal Prediction This book is about conformal prediction and related inferential techniques that build on permutation tests and exchangeability. These techniques are useful in a diverse array of tasks, including hypot...

This new textbook by Anastasios Angelopoulos, Rina Barber and Stephen Bates looks really neat! arxiv.org/abs/2411.11824

now, if we think of p(output | prompt, a few examples) as a predictive distribution p(y|x, D) ... it looks very much like learning to me :)

see e.g. my slide deck on drive.google.com/file/d/1B-Ka...