CMSP Seminar (Atomistic Simulation Webinar Series): Inferring causality in dynamical processes with the Information Imbalance of distance ranks
Starts 21 Jun 2023 13:15
Ends 21 Jun 2023 14:30
Central European Time
Luigi Stasi Seminar Room (and via Zoom)
Vittorio del Tatto
Uncovering causal relationships between time-dependent observables is a problem which goes at the heart of scientific research. However, distinguishing a cause-effect relationship from a correlation, possibly induced by a quantity which is not observed, or unknown, is still considered an open problem, especially if the effect is triggered by the simultaneous variation of a large number of variables. Nowadays, the performance of many causal detection methods degrades quickly with the number of variables that one should consider. Moreover, in real-world contexts most of the approaches lead to false-positives, namely have difficulties in recognizing the absence of causality.
In this talk, I will illustrate a new method built on the Information Imbalance measure, which significantly mitigates these problems. In particular, the use of distance ranks instead of plain variables allows circumventing the problem of estimating probability densities, handling high-dimensional data that are challenging for other approaches.