Copenhagen Causality Lab (CoCaLa)
Causality is a fundamental concept in science, and our research is centered around this concept and how it relates to statistical modeling and data analysis. Some main aspects of our research are outlined below.
Independence statements in probability models can be studied via graphs, and models based on e.g. directed acyclic graphs are used widely as causal models. Graph theory in relation to causal modeling is a core activity of our research. The goal is a deep understanding of how graphical models in a wide sense can be used to encode independence, how they behave as mathematical objects and as statistical models, and how to compute efficiently with graphical models for practical data analysis.
There is a flow from cause to effect along the arrow of time. This is a salient assumption of a causal model, which can be made explicit for models of discrete-time dynamical systems (time series models) using conventional graphical models. For event process models and other continuous-time dynamic stochastic models new developments are required. One goal is to expand the applicability of graphical models beyond the discrete-time framework. Another is to develop practical methods for causal modeling of dynamic stochastic systems.
Data driven selection among multiple causal models is a main practical challenge. We investigate constraint and score based algorithms for graphical model selection. A particular focus is on models with latent variable structures to handle incomplete observations from a causal model. Specific developments include non-parametric conditional independence tests, model scoring, and invariant prediction.
PhD and postdoc positions within CoCaLa will be announced via the departments regular calls, see Vacant positions.
Positions related to specific projects within CoCaLa are described in the general calls, but you are welcome to contact us if you have questions regarding open positions.
We are funded by
- VILLUM FONDEN Young Investigator, Causal Learning in Real World Applications, 2018-2023, PI: Jonas Peters
- Carlsberg Foundation, Learning by doing: How to adapt to changing environments, 2018-2021, PI: Jonas Peters
- VILLUM FONDEN, Causal inference with incomplete data, 2016-2020,
PI: Niels Richard Hansen
- The Dynamical Systems Interdisciplinary Network, 2013-2017,
PI: Susanne Ditlevsen