Leonard Bereska
Leonard Bereska
PhD student, Informatics Institute, University of Amsterdam
- Joined December 2019
Names
Emails
****@hotmail.de (Confirmed)
, ****@zi-mannheim.de (Confirmed)
, ****@uva.nl (Confirmed)
, ****@protonmail.com (Confirmed)
Personal Links
Career & Education History
PhD student
Informatics Institute, University of Amsterdam (uva.nl)
2021 – 2025
Researcher
Theoretical Neuroscience, Central Institute of Mental Health (zi-mannheim.de)
2019 – 2021
MS student
Physics, Heidelberg University (uni-heidelberg.de)
2016 – 2019
Researcher
Physics, Heidelberg University (uni-heidelberg.de)
2012 – 2016
Advisors, Relations & Conflicts
Expertise
ai security, adversarial robustness, interpretability, machine learning theory
2024 – Present
interpretability, developmental interpretability, singular learning theory, neural network dynamics, local learning coefficient
2024 – Present
ai safety, mechanistic interpretability, transformer models, large language models, transparency, interpretability, circuit analysis, causal interpretability
2023 – Present
ai safety, mechanistic interpretability, polysemanticity, superposition, neural network representations, feature disentanglement, sparse autoencoders
2023 – Present