• Home
• Staff & Students
• Vacancies

Jonas Peters

 

Wednesday 28th January 2015

Time: 4pm

 

B10 Basement Floor Seminar Room

Alexandra House, 17 Queen Square, London, WC1N 3AR

 

 

Causal Inference using Invariant Prediction

 

Why are we interested in the causal structure of a data-generating 
process? In a classical regression problem, for example, we include a 
variable into the model if it improves the prediction; it seems that no 
causal knowledge is required. In many situations, however, we are 
interested in the system's behavior under a change of environment, i.e., 
under a different distribution. Here, causal structures become important 
because they are usually considered invariant under those changes and 
can therefore be used to answer questions about this new distribution. 
For example, a causal prediction (which uses only direct causes of the 
target variable as predictors) remains valid even if we intervene on 
predictor variables or change the whole experimental setting. We propose 
a method that exploits the invariance principle when data from different 
environments are available. This talk concentrates on ideas and concepts 
and does not require any prior knowledge about causality.