Nadejda Drenska

Research profile

Research summary

Nadejda Drenska is a researcher at Louisiana State University whose recorded work centers on the mathematics of prediction with expert advice: online-learning problems recast as geometric stopping games and analyzed with partial differential equations. In a 2026 arXiv paper, 'An explicit solution of the five-expert prediction PDE and the exact optimality set of COMB', she is last author; it explicitly solves the five-expert prediction PDE, disproves the claimed optimality of the COMB strategy, and supports the result with exact Bernstein certificates and a Lean 4 formalization, earning Evolve's 'high' reproducibility-readiness rating. A second 2026 preprint, which she first-authors, applies target-aware state-adaptive p-Dirichlet graph neural regression to non-invasive body-composition estimation, extending her range into applied graph machine learning. Bibliometric traces are early-stage but steep: OpenAlex records an h-index of 4 and an i10-index of 1, 20 works over the last three years, and output accelerating from 1-3 works per year between 2019 and 2025 to 16 works in 2026. Topic counts concentrate on advanced bandit algorithms (11) and game theory (10), with stochastic processes (8), optimization and search (7), and formal verification methods (5). Evolve scores activity 89, focus 100, and reproducibility 100; the output and impact axes remain unscored for insufficient data.

Research keywords

Highlights

• Explicit five-expert prediction PDE solution and the exact optimality set of COMB

  In this 2026 arXiv paper (last author), the five-expert prediction PDE is solved explicitly as a geometric stopping game with a viscosity-solution analysis, and the exact optimality set of the COMB strategy is determined — overturning its claimed optimality — with minimax-regret consequences made precise.

• Computer-assisted rigor: Bernstein certificates formalized in Lean 4

  Evolve's review highlights the verification layer of the same paper: exact Bernstein certificates plus a Lean 4 formalization turn the disproof of COMB optimality into a machine-checkable result, underpinning its 'high' reproducibility-readiness rating.

• State-adaptive p-Dirichlet graph neural regression for body composition

  A 2026 first-authored arXiv preprint proposes target-aware, state-adaptive p-Dirichlet graph neural regression for non-invasive body-composition estimation. It is in Evolve's catalog but not yet reviewed, so no verdict, keywords, or reproducibility rating exist for it.

Timeline

1. 2019Earliest year in the OpenAlex counts-by-year record, with 1 work.
2. 20222 works recorded — the highest single-year count up to that point.
3. 2023No works recorded in the counts-by-year stream — the only empty year between 2019 and 2026.
4. 20253 works recorded, immediately preceding the following year's surge.
5. 202616 works recorded, the highest year on file, including both catalog papers: the five-expert prediction PDE solution (last author) and the p-Dirichlet graph neural regression preprint (first author).

Papers in the Evolve catalog

• 2026-09-14

  An explicit solution of the five-expert prediction PDE and the exact optimality set of COMB

Frequent co-authors