🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

Excited to announce that our work with Liudmila Prokhorenkova and Aleksei Ustimenko on uncertainty estimation for tabular data via ensembles of gradient-boosted decision trees (GBDTs) has been accepted to ICLR2021! Paper available here: openreview.net/forum?id=1Jv6b…

Today at #ICLR2021 with @SergeyI49013776 we'll present our work on how to learn on graphs with tabular features: "Boost then Convolve: Gradient Boosting Meets Graph Neural Networks" openreview.net/forum?id=ebS5N… Join us at Poster Session 8: 9 am PDT — 11 am PDT iclr.cc/virtual/2021/p…

#ICLR2021 continues, and today we’ll discuss how to estimate uncertainty of GBDT models: “Uncertainty in Gradient Boosting via Ensembles” openreview.net/forum?id=1Jv6b… Poster Session 11: 9 am PDT — 11 am PDT iclr.cc/virtual/2021/p…

Hooray! Our paper on validation of cluster similarity metrics just accepted for #ICML2021 :) The preprint is available here: arxiv.org/abs/1911.04773 Congrats to my co-authors Liudmila Prokhorenkova and Martijn Gösgens!

1/7 We are happy to launch the Shifts Challenge on uncertainty and robustness to distributional shift NeurIPS Conference! The challenge is jointly organized by Yandex.Research together with the OATML group (OATML_Oxford) the Cambridge University Speech Group. research.yandex.com/shifts

Check out our paper on how to compare classification performance measures. #NeurIPS2021

The #NeurIPS2021 Shifts Challenge on uncertainty and robustness is over! Big thanks to all participants! Come to our mini-workshop on Dec 7th at 18:25 GMT where we will announce the winners, and have 3 keynotes and a panel discussion with amazing speakers! See schedule below:

Tomorrow at #NeurIPS2021 we’ll present our work on how to evaluate classification performance measures: “Good Classification Measures and How to Find Them”. Check out our post and join us at Poster Session 6: 9:30 am — 11 am PDT research.yandex.com/news/how-to-va… neurips.cc/virtual/2021/p…

[1] The Shifts Project is proud to launch the Shifts Challenge 2022, which examines robustness to distributional shift and uncertainty quality in two high-risk tasks of great societal importance. To take part, go to our site on Grand Challenge shifts.grand-challenge.org .

Check out our post on graph-based approaches to nearest neighbor search

Heterophilous graphs, in which edges connect dissimilar nodes, are challenging for GNNs. In our #iclr2023 paper, we propose a new benchmark of 5 structurally diverse heterophilous graph datasets. Read more in this post by Oleg Platonov and Liudmila Prokhorenkova research.yandex.com/blog/introduci…

Testing graph models under diverse and meaningful distributional shifts is crucial for evaluating their robustness and uncertainty. In this post by Gleb Bazhenov, we present a simple approach to create structural shifts in any graph dataset. research.yandex.com/blog/structura…

New blog post! We summarize key findings from our #NeurIPS2023 paper on training neural algorithmic reasoning models without hints. Read how architectural modifications & contrastive learning can match (and outperform) hint-based models on CLRS Benchmark: research.yandex.com/blog/neural-al…