Structural Intervention Distance for Evaluating Causal Graphs

By Jan Peters, Peter Buhlmann. 2015

190 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Statistics and Probability, Causal Inference & Discovery

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Jan Peters
2. Peter Buhlmann
3. Bernhard Schölkopf
4. Simon Lacoste-Julien
5. Marloes H. Maathuis
6. Yoshua Bengio
7. David Lopez-Paz
8. Niklas Pfister

In Statistics and Probability

1. Jan Peters
2. Peter Buhlmann
3. Bernhard Schölkopf
4. Marloes H. Maathuis
5. Markus Kalisch
6. Niklas Pfister
7. Simon Lacoste-Julien
8. Sébastien Lachapelle

In Causal Inference & Discovery

1. Peter Buhlmann
2. Jan Peters
3. Bernhard Schölkopf
4. Simon Lacoste-Julien
5. Marloes H. Maathuis
6. Niklas Pfister
7. Sébastien Lachapelle
8. Philippe Brouillard

Citations in the corpus

• Gradient-Based Neural DAG Learning (2019, methodology): “Introduced by Peters & B¨uhlmann (2015), SID counts the number of interventional distribution of the form p ( x i | do ( x j = ˆ x j )) that would be miscalculated using the parent adjustment formula (Pearl, 2009) if we were to use the predicted DAG instead of the ground truth DAG to form the…”
• Kernel‐based tests for joint independence (2016, methodology): “1000 repetitions): (left) shows how often each method estimated the correct DAG, (right) shows the average structural intervention distance (SID, small is good) between the correct and estimated DAG (Peters and Bühlmann, 2015). 5.3.5 Causal inference (real data example) We now apply the DAG verificat”
• The Case for Evaluating Causal Models Using Interventional Measures and Empirical Data (2019, unknown): “As shown in Figure 4, some of the findings that would be reached with SID and SHD are not supported by a TV evaluation.”
• A Complete Generalized Adjustment Criterion (2015, methodology): “ as done for the generalized back-door criterion for single interventionsby Maathuis and Colombo (2015). In turn, these results could then be used to characterize distances between graphs, as done by Peters and Bühlmann (2015). Future work might also explore under which circumstances our restriction”
• Partition MCMC for Inference on Acyclic Digraphs (2015, background): “In the context of causal graphs, a distance was recently defined (Peters and Bühlmann 2015), though purely based on a graphical criterion, directly related to the network interpretation in terms of differences in the causal inference statements deriving from each structure.”