Dr. Stefan Perko

Research

• Publications

  • Stefan Perko. 2026.
    On the weak convergence of shuffled random walks.
    Accepted in Bernoulli. 
  • Stefan Ankirchner, Sören Christensen, Jan Kallsen, Philip Le Borne, Stefan Perko. 2026.
    Learning to Steer with Brownian Noise.
    Accepted in SIAM Journal on Control and Optimization. [ArXiv]Externer Link
  • Stefan Ankirchner & Stefan Perko. 2023. 
    A comparison of continuous-time approximations to stochastic gradient descent.
    Journal of Machine Learning Research. [PDF]Externer Link
• Preprints

  • Stefan Perko. 2025.
    Towards Continuous-Time Approximations for Stochastic Gradient Descent without Replacement. [ArXiv]Externer Link
  • Stefan Perko. 2023. 
    Unlocking optimal batch size schedules using continuous-time control and perturbation theory. [ArXiv]Externer Link
  • Stefan Ankirchner & Stefan Perko. 2022.
    Towards diffusion approximations for stochastic gradient descent without replacement. [HAL]Externer Link
• Theses

  • Modified Equations for Stochastic Optimization. PhD Thesis. Friechrich Schiller University Jena, 2026.[ArXiv]Externer Link
  • Optimal Control of Stochastic Gradient Descent via Stochastic Modified Equations. Masters thesis. Friechrich-Schiller-Universität Jena, 2020.
• Interests

  • SDE approximations of stochastic optimization algorithms
  • mathematics of (explaining) machine learning
  • Areas: stochastic analysis, functional central limits, stochastic optimization, backward error analysis and modified equations, copulas, permutons and large random permutations, rough paths, learning theory, stochastic control theory, applied category theory