Picture of the author

Jakiw Pidstrigach

PhD Researcher, Potsdam

Interested in generative modeling, markov chain monte carlo, stochastic analysis, interacting particle systems and mean-field analysis.

Google scholar icon

Blog

Tutorial on implementing Diffusion Models from scratch

We derive diffusion models from scratch and relate their discrete time formulation to their continuous time formulation, also called score-based generative models.

Diffusion Models: Continuous time and generalization

We study practical and theoretical aspects of the continuous time formulation of diffusion models. A focus is set on generalization capabilities.

Comparing Diffusion Models to annealead Langevin dynamics

We empirically compare the performance of diffusion models to classical MCMC (Langevin dynamics) by viewing it a simulated annealing scheme for Langevin dynamics.

Conferences

Generative Models for inverse problems deep-dive, Cambridge, 2023

Talk: Infinite-Dimensional Diffusion Models

NeurIPS, New Orleans, 2022

Poster: Score-Based Generative Models Detect Manifolds

Turing Theory and Methodology Challenges Week, London, 2022

Co-organized the "Score-based generative models: Implementation, optimisation, generalisation" challenge.

StatMathAppli, Frejus, 2022

Talk: Score-Generative Models and generalization.

Oberwolfach, 2022

Data Assimilation – Mathematical Foundation and Applications.

Education

Cambridge University, 2023/06

Research visit to the Isaac Newton Institute as part of the Data-Driven Data-Driven Engineering programme.

Massachusetts Institute of Technology, 2022/10 - 2022/12

Research visit to Youssef Marzouk's group. Working on infinite-dimensional diffusion models.

University of Potsdam, 2019 -

Working on my PhD in Sebastian Reich's group and the SFB 1294.

University of Bonn, 2013 - 2019

Masterthesis (supervised by Andreas Eberle): Coupling and convergence of Hamiltonian Monte Carlo on Hilbert spaces.

The story so far

In the beginning the Universe was created. This has made a lot of people very angry and been widely regarded as a bad move.

Publications

See Google Scholar (Link).