Nikita Kiselev

Nikita Kiselev

Personal Website
kiselev.ns@phystech.edu
@kisnikser
Google Scholar
eLibrary

He received his bachelor’s (2024, with honors) and master’s (2026, with honors) degrees from the Intelligent Systems Department at MIPT and is now a PhD student at the department, advised by Andrey Grabovoy. His PhD thesis is on scaling parametric machine learning models based on the local geometry of the optimization surface. He is a Senior ML Research Engineer at Kandinsky Lab, Sber, where he works on post-training (SFT) and alignment (RLHF) of Kandinsky 6.0 text- and image-to-audio-video models; before that he worked on Kandinsky 5.0 and 4.1 image and video models. He organizes and teaches the Deep Learning course at MIPT and teaches Diffusion Models at Sber University. He is an author of more than 10 research papers.

Professional interests: generative AI (diffusion models, alignment), loss landscape geometry, Hessian analysis, sample size determination, optimization.

Publications

Conference papers

  1. D. Yarmoshik, A. Rogozin, N. Kiselev, D. Dorin, A. Gasnikov, D. Kovalev. Decentralized Optimization with Coupled Constraints // ICLR, 2025.
  2. V. Meshkov, N. Kiselev, A. Grabovoy. ConvNets Landscape Convergence: Hessian-Based Analysis of Matricized Networks // Ivannikov ISPRAS Open Conference, 2024.

Journal papers

  1. N. Kiselev, V. Meshkov, A. Grabovoy. Robust Convergence of Loss Landscapes through Distributional Averaging // Proceedings of the Institute for System Programming of the RAS, 2026.
  2. D. Dorin, N. Kiselev, A. Grabovoy. Decoding visual information from neural signals: image reconstruction based on joint functional magnetic resonance imaging and electroencephalography analysis // Informatics and Applications, 2026.
  3. N. Kiselev, A. Grabovoy. Sample Size Determination: Posterior Distributions Proximity // Computational Management Science, 2025.
  4. N. Kiselev, A. Grabovoy. Sample Size Determination: Likelihood Bootstrapping // Computational Mathematics and Mathematical Physics, 2025.
  5. N. Kiselev, A. Grabovoy. Unraveling the Hessian: A Key to Smooth Convergence in Loss Function Landscapes // Doklady Mathematics, 2024.
  6. D. Dorin, N. Kiselev, A. Grabovoy, V. Strijov. Forecasting fMRI Images From Video Sequences: Linear Model Analysis // Health Information Science and Systems, 2024.

Technical reports

  1. Kandinsky Lab. Kandinsky 6.0 Video: Foundation Models for Synchronized Video and Audio Generation // arXiv, 2026.
  2. Kandinsky Lab. Kandinsky 5.0: A Family of Foundation Models for Image and Video Generation // arXiv, 2025.

Preprints

  1. E. Petrov, N. Kiselev, V. Meshkov, A. Grabovoy. Closing the Curvature Gap: Full Transformer Hessians // arXiv, 2025.
  2. N. Kiselev, A. Grabovoy. Curvature-Aligned Probing for Local Loss-Landscape Stabilization // arXiv, 2026.

Talks and conference abstracts: kisnikser.github.io.

Courses

My First Scientific Paper Computer Vision Deep Learning