Hi, I'm Andrey Veprikov 👋

AI & Optimization Researcher · PhD Student at MIPT

---

I research optimization for deep learning — how to train and adapt large models faster, cheaper, and more reliably. I am a PhD student at MIPT who combines academic research (15+ papers, including top-tier venues) with applied AI in industry, where I lead a practical-AI direction at Sber, Russia's largest bank.

🔬 Research Interests

• Optimization for ML — next-generation optimizers (Muon, SignSGD), zero-order / gradient-free methods, warm-up scheduling
• Efficient LLMs — parameter-efficient fine-tuning (LoRA / PEFT), pretraining, quantization & inference acceleration
• Distributed & federated optimization — decentralized methods, Markovian compression, trusted AI
• AI Scientist — autonomous agents for conducting scientific research

📝 Selected Publications

• Softsign: Smooth Sign in Your Optimizer for Better Heterogeneity Handling — ICML 2026 (A*) · arXiv
• WeightLoRA: Keep Only Necessary Adapters — ACL 2026 (A*) · arXiv
• Methods for Optimization Problems with Markovian Stochasticity and Non-Euclidean Geometry — AAAI 2026 (A*) · arXiv
• Markovian Compression: Looking to the Past Helps Accelerate the Future — UAI 2026 (A) · arXiv
• New Aspects of Black-Box Conditional Gradient — Chaos, Solitons & Fractals (Q1)

📚 Full list with metrics on Google Scholar — 50+ citations, h-index 5.

🛠️ Open Source

• LIB — a benchmarking framework for first-order optimization methods across CV, NLP and LLM fine-tuning
• brain-lab-research — I administer the lab's GitHub org and publish reproduction code for our papers
• llm-baselines — contributions to a fork of EPFL's pretraining codebase: new-generation optimizers and benchmarks
• claude-brainlab — 50+ skills & agents that automate the research cycle (paper ingestion, reviewing, writing)
• Active contributor to the Intelligent Systems Dept. org (290+ repositories), home of the most-starred Relaxit

💼 Experience & 🎓 Education

• Sber (Russia's #1 bank) — Head of an applied-AI direction, from intern to head in under a year
• MIPT / BRAIn Lab — research engineer; optimization for LLM training, next-gen optimizers, AI Scientist
• Research roles at ISP RAS, IITP RAS, and MBZUAI (Abu Dhabi) — international collaborations
• MIPT — M.Sc. with honors (GPA 9.5 / 10), now a PhD student

🧑‍⚖️ Academic Service

Golden Reviewer at ICML · reviewer for NeurIPS and TMLR

📫 Get in Touch

Email · Telegram · Google Scholar · Full CV in English / Русский