Theses
Open Bachelor's and Master's thesis topics
Thesis Supervision
I am open to supervising Bachelor’s and Master’s theses related to optimization, dynamical systems, learning theory, game theory, stochastic algorithms, and formalization of mathematical methods.
Bachelor’s theses are usually suitable for focused literature reviews, numerical experiments, or simplified models. Master’s theses can be more research-oriented and may involve theoretical analysis, algorithm development, formalization, or more substantial computational work.
If you are interested in one of the topics below, please contact me with a short description of your background, your degree program, and the proposal you are interested in.
Open Thesis Proposals
Non-Euclidean Geometry and Stability of Player-Wise Mirror Learning in Games
Level: Master’s thesis
Duration: 6 months
Type: Research thesis with theorem, counterexample, and worked examples
Status: Open
This topic studies which Hilbert-space stability mechanisms for player-wise learning in games survive under non-Hilbert Banach geometry, and which fail. The thesis combines geometry-dependent stability in simple games, explicit failure modes of Euclidean proof arguments, and a positive relative-monotonicity principle.
AI-Assisted Formalization of Finite Learning in Games in Lean 4
Level: Master’s thesis
Duration: 6 months
Type: Library-building thesis with a fully verified core theorem
Status: Open
This topic builds a reusable Lean 4 library for finite-game foundations of learning in games. The central formalized theorem is the classical link between no-regret learning and approximate coarse correlated equilibrium in a finite normal-form game.
AI-Assisted Formalization of Bregman Geometry and Mirror Learning in Lean 4
Level: Master’s thesis
Duration: 6 months
Type: Library-building thesis with a fully verified core theorem
Status: Open
This topic develops a Lean 4 library for Bregman geometry, variational inequalities, monotone operators, mirror steps, and mirror-descent regret or Lyapunov bounds. The goal is a precise formal layer that future projects in verified learning in games can build on.