Head of AGWI

Gordei Verbii

Reframing AI problems through deep mathematics

Gordei leads AGWI, reframing AI problems through deep mathematics. He is a PhD student at Shanghai Jiao Tong University, supervised by Yaoyu Zhang, researching the theory of deep learning. His work connects probability, geometry, optimal transport, and control.

GitHub ↗Google Scholar ↗Yaoyu Zhang ↗

Research interests

Theory of deep learning
Optimal transport
Geometry
Probability
Control

Education

PhD, Shanghai Jiao Tong University — in progress; theory of deep learning, supervised by Yaoyu Zhang.
Specialist degree in Fundamental Mathematics, Lomonosov Moscow State University, 2020–2026. Concentration in probability theory; minor in optimal control.

Selected experience

Head of AGWI, connecting mathematical foundations with AI research.
Lead AI Researcher at Stratwise since September 2025, working across ranking, reinforcement learning, sequence models, and evaluation.
Research in optimal transport at the Vega Institute Foundation, 2024–2025, and reinforcement learning on Cayley graphs through the ML in Algebra project.
Research supervision and mentoring since April 2025; assistant scientific supervision at Lomonosov MSU since September 2024.

RESEARCH RECORD

Selected & shared
work.

2026Preprint

The Distributional View of Knowledge Distillation

Gordei Verbii, Juho Lee

arXiv:2608.15215

Gordei Verbii

Research interests

Education

Selected experience

Selected & sharedwork.

The Distributional View of Knowledge Distillation

One Inverse Step is a Convex Program: Bayes-Limit Calibration of Diffusion Inversion

Rough Paths to Certified Reliability

Domain-constrained Knowledge Accumulates Flashes of Insight through Contradictions at the Border of Domains

FiGLi: Finsler Gauge on Lie Algebra for Guiding Vector Fields

CayleyPy RL: Pathfinding and Reinforcement Learning on Cayley Graphs

Cohomological Consistency in Modular Soft Robot Networks

Spectral Fingerprinting of Deformable Bodies: Laplace–Beltrami Invariants for Soft Robot State Estimation

Persistent Homology for Topological Classification of Soft Robot Deformation Modes

Microlocal Observability: Wave Front Analysis for Vibration Control in Soft Structures

Information-Geometric Control: Natural Gradient Methods for Adaptive Soft Robot Learning

Physics-Informed Neural Operators for Closed-Loop Magnetic Swarm Control of Soft Robots

Multi-Fidelity Robotic Co-Design via the Refinement Limit

Hidden Markov Models with Memory and Multi-Criteria Queue Optimization for Mining Fleet Management

The Stiefel–Procrustes Bridge: Differential Geometric Framework for Smooth Robot Locomotion–Manipulation Transitions

Swarm Control via ODE-Driven Optimal Transport–Based Renormalization Group Flow

Voronoi-Based Multi-Robot Autonomous Exploration in Unknown Environments via Randomized SOM-Based Q-Learning

Optimal Stopping via Martingale Rainbow Deep Q-Learning: Risk-Sensitive Financial Decision-Making

Self-Organizing Maps for Spatiotemporal Analysis of SSVI Vol. Surface Regimes

Optimal Recovery Meets Bayesian Learning: Where Worst-Case Bounds Pay Off

Research projects

SSVI Surface Calibration via Schrödinger Bridge ↗

Reinforcement Learning for Optimal Stopping with Financial Applications ↗

Scale vs. Quantization: Mapping the Memory Wall on Consumer Hardware ↗

Actuarial-Informed Neural Networks ↗

Selected & shared
work.