Ivan Rubachev

This account is verified. Learn more

@puhsuuu

ML Researcher @YandexResearch | Tabular ML

F#1M/3146 SAE feature puhsu.net Joined August 2016

2.1K Following 542 Followers 400 Posts

Latest posts 5

Pinned post

Ivan Rubachev @puhsuuu Oct 23

I've long argued for more diverse evaluation setups in tabular ML research. This will continue. It was a key motivation behind our new paper

A thread on data scaling in MLPs, GBDTs and tabular foundation models, and what it means for generalization.

arxiv.org/abs/2610.26794

Ivan Rubachev @puhsuuu Oct 30

TabPack is great! now it has а package

Yura Gorishniy @YuraFiveTwo Oct 30

Double announcement!

🤖 TabPack: an effective, practical baseline for medium-to-large tabular datasets. Competitive out of the box on benchmarks and trainable on a modern MacBook

📦 TabPackage: a Python package for training TabPack and building similar models

👇 Links & Thread

View quoted post on X

Ivan Rubachev @puhsuuu Oct 20

Me reading arxiv.org/abs/2403.03950 (by @JesseFarebro, @pcastr, @agarwl_ et al) thinking about a small tabular ML regression project

Turned out to be bigger than expected, very pleased the paper is out, its an extensive exploration of regression objectives and then some

Yandex Research @YandexResearch Oct 20

Training neural networks for tabular regression? Our latest work may be of interest

And it's not only about which of the 18! regression objectives is best, but touches on the role of noise and global early stopping in tabular DL in general
arxiv.org/abs/2610.13909

Thread (1/N)
View quoted post on X

Ivan Rubachev @puhsuuu Sep 30

I'm a fan of this!

Been watching and rooting for the guys for a long time. They are pursuing an interesting direction in agents.

Going lower into the inference runtime, for making LLM agents more asynchronous, I think we need more of this

Yandex Research @YandexResearch Sep 30

LLMs are general asynchronous agents

We're releasing AsyncLLM - a framework for pretrained LLMs to observe, reason and act concurrently, with no additional training

Paper: huggingface.co/papers/2609.35…
Code: github.com/dvmazur/async_…
Website with Demos: dvmazur.github.io/async_llm

🧵...
View quoted post on X

Ivan Rubachev @puhsuuu Sep 29

I think the library released github.com/NVIDIA/structu… is very clean in general: stuff like preprocessing, tabular tensor abstraction and API design, besides the FM stuff announced in the tweet. Tabular ICL foundation models are also progressing surprisingly quickly!

github.com GitHub - NVIDIA/structured-data-models: Foundation Models for Structured Data Foundation Models for Structured Data. Contribute to NVIDIA/structured-data-models development by creating an account on GitHub.

Jure Leskovec Verified account @jure Sep 29

I’m very excited to share @nvidia Kumo Tabular, a new family of foundation models for tabular data.

Kumo Tabular establishes the new Pareto frontier across the entire accuracy–inference-time tradeoff.

Just as importantly, we are releasing it openly: open weights, open-source software, and a permissive license for commercial use.

HuggingFace: huggingface.co/nvidia/Kumo-Ta…
GitHub: github.com/NVIDIA/structu…

View quoted post on X