Big maths and I.
I research methods and limitations of causal inference and their intersection with machine learning.
I believe that causal inference can work in the real world only if we are honest about its assumptions. That is why I am looking at problem settings with trustworthy properties:
- Partial identification is a very attractive alternative to full identification due to weaker assumptions.
- New directions such as Causal Bayesian Optimisation or Synthetic Control will play a crucial role in the future.
- I also find topological perspectives on causal inference particularly useful for understanding limits of causal inference.
Message me at email@example.com
|Jan 12, 2023|| Two papers accepted at CLeaR (Causal Learning and Reasoning) 2023 |
Non-parametric identifiability and sensitivity analysis of synthetic control models (with Spotify)
Stochastic Causal Programming for Bounding Treatment Effects
|Oct 22, 2022||New page, new life. Moving over from https://jakobzeitler.weebly.com|
|Oct 19, 2022||Presented our recent work on partial identification at the Institute for Data Science and Artificial Intelligence. Recording here.|
|Sep 19, 2022||Completed my summer intership at Spotify’s new Advanced Causal Inference lab. Together with Ciaran Lee I investigated the assumptions of synthetic control, hoping to share results on that beginning 2023.|
|Oct 20, 2019||Started my PhD with Ricardo Silva at the new Centre for Doctoral Training in Foundational Artificial Intelligence. Thanks to UKRI and DeepMind for the generous funding.|
- The Causal Marginal Polytope for Bounding Treatment EffectsarXiv preprint arXiv:2202.13851 2022
- Stochastic Causal Programming for Bounding Treatment EffectsarXiv preprint arXiv:2202.10806 2022
- Algorithmic recourse in partially and fully confounded settings through bounding counterfactual effectsarXiv preprint arXiv:2106.11849 2021