About Me

I am a Ph.D. student at the Department of Statistics, University of Michigan and am chiefly interested in generative AI, with a particular emphasis on its interpretability and scientific applications.

Previously, I did research at UC Berkeley, where I was part of Project CETI, working on furthering our understanding of sperm whale communication using modern machine learning.

I also hold an advanced degree in Mathematical Physics from the University of Ljubljana, Slovenia, and worked in both the technology and financial industries in the time between my Physics and Statistics degrees.

Interests
  • Generative AI
  • AI for Science
  • Interpretable Machine Learning
Education

  • M.A. Statistics, 2022

    University of California, Berkeley

  • M.Sc. (equivalent) Mathematical Physics, 2016

    University of Ljubljana, Slovenia

Recent Publications

Andrej Leban (2025). Energy-Tweedie: Score meets Score, Energy meets Energy. On ArXiv.

PDF Cite arXiv alphaXiv

Andrej Leban (2025). Distributional Autoencoders Know the Score. NeurIPS 2025.

PDF Cite Code Poster Slides NeurIPS page OpenReview arXiv alphaXiv

Gašper Beguš, Ronald L. Sprouse, Andrej Leban, Miles Silva, Shane Gero (2023). Vowel- and Diphthong-Like Spectral Patterns in Sperm Whale Codas. Open Mind (MIT Press).

PDF Cite Open Mind

Gašper Beguš, Andrej Leban, Shane Gero (2023). Approaching an unknown communication system by latent space exploration and causal inference. On ArXiv.

PDF Cite Code arXiv

Marco Miccheli, Andrej Leban, Andrea Tacchella, Andrea Zaccaria, Dario Mazzilli, Sébastien Bratières (2022). A Bayesian approach to translators' reliability assessment. On arXiv.

PDF Cite arXiv

See all publications

Popular Topics

AI4Science Animal Communication Autoencoders Bash C++ Causal Inference Data Science Deep Learning Denoising Diffusion Diffusion Models Energy-Based Models Generative Models GitHub Machine Learning Manifold Learning Monte Carlo Score Matching Statistics Unsupervised Learning