Causal ML Lab

We develop new machine learning methods for causal inference and decision-making.

Our vision

Existing machine learning (ML) methods typically model correlations in the data, but not causal relationships. Such correlation-based methods can have suboptimal performance and even lead to systematic errors. To address this, we aim to develop novel ML models for causal inference, i.e., that are able to infer causal effects from observational data. Concretely, we advance causal ML toward the directions of flexibility, efficiency, and robustness:

  • (1) Flexibility: We develop general-purpose causal ML models that can handle a broad range of settings, including high-dimensional, time-series, and multi-modal data.
  • (2) Efficiency: We develop efficient learning algorithms (e.g., based on statistical theory) that we carefully incorporate into our models.
  • (3) Robustness: We also develop new techniques for verifying and improving the robustness of causal ML, e.g., methods that operate under violation of standard assumptions or methods that account for data/estimation uncertainty

Moreover, we aim to achieve better visibility and further applicability of our research through principles of (i) Open-source development and (ii) Real-world applications.

  • (i) Open-source development: We fill white spots in the causal ML toolchain to improve industry uptake. An open-source code thus accompanies every project. Also, we release the pre-trained models and design guidlines to adapt any method for a problem at hand.
  • (ii) Real-world applications: We demonstrate performance gains of newly developed causal ML methods through real-world case studies from business, public policy, and medicine.

Our Lab

Co-directors

Dennis Frauen, M.Sc.

Doktorand und Wissenschaftlicher Mitarbeiter

Valentyn Melnychuk, M.Sc.

Doktorand und Wissenschaftlicher Mitarbeiter

Our research

We seek to publish in the top-tier outlets of ML research (e.g., NeurIPS, ICML, ICLR) across different directions:

(1.1) Causal ML for multi-modal data

(1.2) Causal ML for time series data

(1.3) Constrained causal ML:

(1.4) Causal ML for heterogeneous data sources:

Future directions: (1.5) Bandits and reinforcement learning

With financial support from the German Federal Ministry of Education and Research as part of the Grant “CausalNet”