Optimization Theory
Convergence analysis for minimax and LMO-based methods, including Frank-Wolfe variants and optimizers used in deep learning.
Igor Ignashin
Research Lead in Machine Learning and Optimization
I am an M.S. candidate at the Moscow Institute of Physics and Technology (MIPT), Faculty of Applied Mathematics and Informatics, Department of Intelligent Data Analysis.
My work focuses on machine learning optimization, stochastic dynamics of SGD, LLM post-training, SFT and teacher distillation, inference acceleration, minimax optimization, and applied traffic assignment.
I work as a Research Lead at BRAIn Lab / MIRAI under the supervision of Alexander Beznosikov, collaborate with Demyan Yarmoshik at LAB MMO in Alexander Gasnikov's research group, and recently worked as a visiting research student at MBZUAI under Eduard Gorbunov.
I also collaborate on SGD analysis and multi-agent reinforcement learning with Andrei Leonidov's team.
Research Focus
Optimization, learning dynamics, and reliable ML systems
Stochastic Dynamics
Experiments and theory for finite-step SGD dynamics beyond Brownian-motion approximations and standard Langevin models.
LLM Efficiency
Post-training and efficiency projects on SFT, teacher distillation, pruning, early exit, multi-agent RL, and LLM training dynamics.
Selected Publications
Recent papers and preprints
Aligning Distributionally Robust Optimization with Practical Deep Learning Needs
Accepted to a NeurIPS 2025 workshop.
Why SGD is not Brownian Motion: A New Perspective on Stochastic Dynamics
NeurIPS 2026 submission on stochastic dynamics of SGD.
Frank-Wolfe Modifications for Equilibrium Traffic Assignment
Published in Computer Research and Modeling.
Stochastic Origin Frank-Wolfe for Traffic Assignment
Presented at TFN-2025 and accepted to the Journal of Mathematical Sciences, Series B.
Modeling Skiers Flows via Wardrop Equilibrium in Closed Capacitated Networks
Presented at TFN-2025 and accepted to the Journal of Mathematical Sciences, Series B.
Conjugate Frank-Wolfe in Machine Learning
Presented at OPTIMA and accepted to CCIS.
Talks and Media
Public talks, programs, and media mentions
Efficient approaches to compressing large language models
Public lecture on distillation, structured pruning, and early exit for large language models.
LLM compression at the AIRI summer school in Tomsk
RIA Tomsk quoted my explanation of layer removal for making large language models smaller while keeping useful quality.
Optimization dynamics and traffic-flow papers
Intelligent Systems at Phystech highlighted my work on SGD dynamics and traffic-flow optimization in its yearly research review.
Stochastic Origin Frank-Wolfe for Traffic Assignment
Presentation at the Traffic Flows on Networks conference at the Sirius Mathematics Center.
Synthetic data for improving object-detection models
Abstract in the proceedings of the 67th MIPT conference on applied mathematics and computer science.
Equilibrium traffic assignment problem
Talk and abstract at the 66th MIPT conference on Frank-Wolfe modifications for equilibrium transportation-flow assignment.
Projects
My work and team research
My projects
Research and engineering projects where I am the main author or a direct contributor.
BRAIn Lab team projects
Projects I lead or supervise with student teams; public links are shown once the repository is ready.
CV
Experience and education
MIPT, Applied Mathematics and Informatics
B.S. in Applied Mathematics and Physics; currently an M.S. candidate in the Department of Intelligent Data Analysis.
Research and industry
Research Lead at BRAIn Lab / MIRAI under Alexander Beznosikov; collaboration with Demyan Yarmoshik and Alexander Gasnikov's LAB MMO; Visiting Research Student at MBZUAI under Eduard Gorbunov; current work on LLM post-training and inference acceleration, including Qwen/DeepScaleR teacher-SFT, RLVR/GRPO-style math training, reward parsing, benchmark reporting, and vLLM early-exit/adaptive decoding pipelines; work on SGD analysis and multi-agent reinforcement learning with Andrei Leonidov's team; former Data Analyst Intern at Yandex.
Thesis repositories
School olympiads
- Winner, Phystech Mathematics Olympiad.
- Winner, Step into the Future Olympiads in Mathematics and Physics.
- Winner, KFU Interregional Mathematics Olympiad.
- Winner, municipal stage of the All-Russian School Olympiad in Mathematics.
Technical stack
Python, C++, SQL, PyTorch, JAX, vLLM, TRL, Hugging Face workflows, SFT, teacher distillation, RLVR/GRPO-style training, reinforcement learning, LLMs, convex and nonconvex optimization, stochastic processes, multi-GPU server workflows, Linux, Git, YQL, DataLens, and Nirvana.