Michael Diskin

Industry. Head of LLM R&D at Wildberries — Russia’s largest e-commerce platform. Built the LLM & embeddings organization from scratch (30+ people, 4–5 teams), shipping search, retrieval, machine translation, and RAG systems at scale. Previously ML at Yandex.

Research. Published at NeurIPS, ICML, ICLR, and EMNLP (700+ citations). Core topics: distributed training, collaborative deep learning, graph neural networks. Co-created Hivemind — an open-source framework for decentralized training.

Teaching. Lecturer at Harbour.Space University and Yandex School of Data Analysis (NLP, Deep Vision & Graphics, Reinforcement Learning).

news

Apr 23, 2026	New paper: Learning When to Be Sparse: Adaptive Activations via Two-Parameter Entropy appeared on OpenReview (Sci4DL workshop @ ICLR 2026). Also arrived in Rio de Janeiro for the conference, happy to connect on site.
Apr 09, 2026	Talk video from Data Fusion 2026: “Do You Really Need a Multi-Agent System? Lessons from Teams That Already Tried.”
Feb 13, 2026	Attended MLWS @ MBZUAI; no talk this time, but many valuable conversations and networking with the community.
Nov 09, 2025	Attended EMNLP 2025, presented at a workshop, and had many productive discussions with colleagues; also helped organize and connect the Russian-speaking NLP/ML community on site.
Nov 01, 2025	Synthetic Proofs with Tool-Integrated Reasoning: Contrastive Alignment for LLM Mathematics with Lean appeared in ACL Anthology (MathNLP @ EMNLP 2025).

selected publications

ICML

SWARM Parallelism: Training Large Models Can Be Surprisingly Communication-Efficient

Max Ryabinin, Tim Dettmers, Michael Diskin, and 1 more author

In International Conference on Machine Learning (ICML), 2023

arXiv HTML PDF

@inproceedings{ryabinin2023swarm,
  title = {SWARM Parallelism: Training Large Models Can Be Surprisingly Communication-Efficient},
  author = {Ryabinin, Max and Dettmers, Tim and Diskin, Michael and Borzunov, Alexander},
  booktitle = {International Conference on Machine Learning (ICML)},
  year = {2023},
  url = {https://arxiv.org/abs/2301.11913},
}

ICLR

A Critical Look at the Evaluation of GNNs under Heterophily: Are We Really Making Progress?

Oleg Platonov, Denis Kuznedelev, Michael Diskin, and 2 more authors

In International Conference on Learning Representations (ICLR), 2023

arXiv HTML PDF Code

@inproceedings{platonov2023critical,
  title = {A Critical Look at the Evaluation of GNNs under Heterophily: Are We Really Making Progress?},
  author = {Platonov, Oleg and Kuznedelev, Denis and Diskin, Michael and Babenko, Artem and Prokhorenkova, Liudmila},
  booktitle = {International Conference on Learning Representations (ICLR)},
  year = {2023},
  url = {https://arxiv.org/abs/2302.11640},
}

NeurIPS

Distributed Deep Learning in Open Collaborations

Michael Diskin, Alexey Bukhtiyarov, Max Ryabinin, and 13 more authors

In Advances in Neural Information Processing Systems (NeurIPS), 2021

arXiv HTML PDF Code

@inproceedings{diskin2021distributed,
  title = {Distributed Deep Learning in Open Collaborations},
  author = {Diskin, Michael and Bukhtiyarov, Alexey and Ryabinin, Max and Saulnier, Lucile and Lhoest, Quentin and Sinitsin, Anton and Popov, Dmitry and Pyrkin, Dmitry and Kashirin, Maxim and Borzunov, Alexander and Villanova del Moral, Albert and Mazur, Denis and Kobelev, Ilia and Jernite, Yacine and Wolf, Thomas and Pekhimenko, Gennady},
  booktitle = {Advances in Neural Information Processing Systems (NeurIPS)},
  year = {2021},
  url = {https://arxiv.org/abs/2106.10207},
}