| CARVIEW |
Select Language
HTTP/2 200
server: GitHub.com
content-type: text/html; charset=utf-8
last-modified: Mon, 24 Nov 2025 07:49:34 GMT
access-control-allow-origin: *
strict-transport-security: max-age=31556952
etag: W/"69240e0e-a13e"
expires: Tue, 30 Dec 2025 09:31:31 GMT
cache-control: max-age=600
content-encoding: gzip
x-proxy-cache: MISS
x-github-request-id: DD28:1387E:9CE363:B05AB7:6953999B
accept-ranges: bytes
age: 0
date: Tue, 30 Dec 2025 09:21:32 GMT
via: 1.1 varnish
x-served-by: cache-bom-vanm7210026-BOM
x-cache: MISS
x-cache-hits: 0
x-timer: S1767086492.821771,VS0,VE203
vary: Accept-Encoding
x-fastly-request-id: 3f16b32c7c9081afd5ed14e03e2db4bb7ff2a153
content-length: 9419
Armin Kekić
I’m a PhD Student at the Max Planck Institute for Intelligent Systems, specializing in integrating causal inference and reasoning into machine learning algorithms.
Prior to my PhD, I worked as an applied scientist at Zalando, where I developed machine-learning-based demand forecast models for price optimization.
My academic background includes studies in physics and applied mathematics at Heidelberg, Oxford, and Paris, with a specific focus on theoretical quantum dynamics and simulation methods.
On this website, I share projects that I work on.
Interests
- Machine Learning
- Causality
- Time Series Forecasting
- Network Science
Education
-
PhD in Machine Learning, ongoing
Max Planck Institute for Intelligent Systems
-
MSc in Applied Mathematics, 2016
University of Oxford
-
BSc in Physics, 2015
University of Heidelberg
News
- Jul. 2025: We won the best paper award for Learning Nonlinear Causal Reductions to Explain Reinforcement Learning Policies at the Workshop on Causal Abstractions and Representations at UAI 2025.
- May 2025: Our paper on learning joint interventional effects from interventions on single variables was accepted at ICML 2025.
- Dec. 2024: I am attending NeurIPS 2024 in Vancouver. Feel free to contact me if you’d like to chat about causality in general or its applications to simulations, time series or representation learning.
- Sep. 2024: Our preprint on estimating joint interventional distributions from marginal interventional data is now available on arXiv.
- May 2024: Our paper on Tageted Reduction of Causal Models was accepted as an oral at UAI 2024.
- Mar. 2024: Our paper on Tageted Reduction of Causal Models was accepted at the AI4DifferentialEquations workshop at ICLR 2024.
- Sep. 2023: Our papers on Causal Component Analysis and multi-environment causal representation learning were accepted at NeurIPS 2023.
- Jul. 2023: Our paper on understanding atomic spectra using network science was published in the Journal of Physics: Complexity.
Recent Publications
Quickly discover relevant content by filtering publications.
Armin Kekić,
Jan Schneider,
Dieter Büchler,
Bernhard Schölkopf,
Michele Besserve
(2025).
Learning Nonlinear Causal Reductions to Explain Reinforcement Learning Policies.
arXiv.
Armin Kekić,
Sergio Hernan Garrido Mejia,
Bernhard Schölkopf
(2025).
Learning Joint Interventional Effects from Single-Variable Interventions in Additive Models.
ICML.
Sergio Hernan Garrido Mejia,
Elke Kirschbaum,
Armin Kekić,
Atalanti Mastakouri
(2024).
Estimating joint interventional distributions from marginal interventional data.
arXiv.
Armin Kekić,
Bernhard Schölkopf,
Michel Besserve
(2024).
Targeted Reduction of Causal Models.
UAI.
Wendong Liang,
Armin Kekić,
Julius von Kügelgen,
Simon Buchholz,
Michel Besserve,
Luigi Gresele,
Bernhard Schölkopf
(2023).
Causal Component Analysis.
NeurIPS.
Experience
PhD Student (Machine Learning and Causality)
Sep 2021 –
Present
Tübingen, Germany
My main interest lies in developing methods for causal representation learning in realistic scenarios. I am a member of the Empirical Inference Department supervised by Bernhard Schölkopf.
Applied Scientist (Algorithmic Pricing)
Feb 2018 –
Aug 2021
Berlin, Germany
At the Pricing and Forecasting Department, our main mission was to develop an automated desicion making system that selects optimal dynamic prices for fashion articles (millions of pricing decisions at each iteration). In particular, I modeled high-dimensional time series using deep learning to predict how price changes affect sales. To make good and reliable decisions in the real world, automated systems have to understand the difference between correlation and causation; this got me intersted in the topic of my PhD.
Researcher (Spectroscopic Networks)
Mar 2017 –
Jan 2018
Heidelberg, Germany
We applied methods from network science to spectroscopic data of atoms and found that we can predict the existence of atomic transitions. Additionally, community structure in spectroscopic networks corresponds to physical properties of the quantum states. This project at the intersection of physics and computer science tried to explore what we can learn about physical systems by purely looking at data science methods, rather than building a microscopic physical model.