Arman Bolatov

ELLIS PhD student at CISPA, advised by Sebastian Stich.

profile.jpeg

Saarbrücken, Germany

My research spans optimization, (machine, deep, federated) learning, efficient inference and training, and learning theory. I did my MSc in Machine Learning at MBZUAI, advised by Martin Takáč, and my BSc in Computer Science at Nazarbayev University.

I’ve also built RAG pipelines and deployed real-time CV models. On the side, I contributed to Beyond Curriculum, a Kazakhstani non-profit running Scoreboard, Blog, and Ask. I also taught olympiad math to high schoolers, who have since gone on to Columbia, Yonsei, and Nazarbayev University, and to medals at the Kazakhstan National Math Olympiad.

Outside of work, I love music, playing guitar, reading history, playing videogames, and I’m an average gym goer (I bench 100kg btw). Always happy to chat! 😊

selected publications

  1. CPAL
    Byzantine-Robust Optimization under (L_0,L_1)-Smoothness
    Arman Bolatov, Samuel Horváth, Martin Takáč, and 1 more author
    In Conference on Parsimony and Learning, 23–26 mar 2026
  2. Preconditioned Norms: A Unified Framework for Steepest Descent, Quasi-Newton and Adaptive Methods
    Andrey Veprikov, Arman Bolatov, Samuel Horváth, and 3 more authors
    arXiv preprint, 2025
  3. ECAI
    Long-Tail Theory under Gaussian Mixtures
    Arman Bolatov, Maxat Tezekbayev, Igor Melnykov, and 3 more authors
    In European Conference on Artificial Intelligence (ECAI), 2023