Vergangene Events

November Special: Causality in Science and Machine Learning

Dieses Meetup liegt in der Vergangenheit

251 Personen haben teilgenommen

Bild des Veranstaltungsortes

Details

*Room changed: C.A.R.L. H03 (480 seats).*

Three experts share with us how new methods make it possible to explore causal structures from data. Sound methods for causal analysis are young and widely unknown.

Introduction: History of causality research and a brief overview of the methods that Judea Pearl presents in "The Book of Why".

Prof. Jörg Breitung, University Cologne: "Causal Analysis in Empirical Economics - attribution of influencing factors in time series".

Dr. Frank Buckler, Founder & CEO Success-Drivers: "The Search for Success Drivers: How machine learning helps to explore root causes of success - for business strategies, other complex systems or even soccer."

Prof. Boudewijn van Dongen, TU Eindhoven: "Process Analytics: Data Analytics driven by Causality. Mining real world data to find networks that model the causal behavior of multi-faceted complex interactions."

Panel debate moderated by Eckhard Siegmann: "How can causal methods find their way into education, science and business?"

18:30 Meet and greet, Foyer C.A.R.L.
18:45 Introduction and talks
20:05 Break
20:20 Q&A, Panel debate
21:00 Closing

We have reserved a room in the student pub 'Labyrinth', Pontstr. 156/158 Aachen for a drink or dinner afterwards.

REGISTRATION: If you do not like to register via this website, notify us about your attendance via kaminski(at)time.rwth-aachen.de
Thank you.

*Abstracts*

Prof. Breitung: "Causal Analysis in Empirical Economics"

Today, two different concepts are typically used for the causal analysis of economic data: (1) Granger causality tests are based on the notion that knowing the cause should help to forecast the event. Accordingly, Granger causality boils down to testing the predictive ability of some variable that is considered as a (prima facie) cause of the event. (2) The microeconometric approach analyses the data as if it was generated by a controlled experiment. Accordingly, this concept adapts statistical methods from analysing clinical trials. After an introduction to both concepts merits and drawbacks are discussed.

Dr. Frank Buckler: "The Search for Success Drivers: How machine learning helps to explore root causes of success"

Existing causal structure learning approaches have several deficiencies that can be overcome by Universal Structure Modeling. It will be explained how this method works and why and when it is possible to conclude causality. Some application examples highlight why Universal Structure Modeling is so useful and relevant today.

Prof. Boudewijn van Dongen: "Process Analytics: Data Analytics driven by Causality"

Each time two or more activities are performed to reach a certain goal, fundamental principles of processes apply. A key concept in processes is causality. One activity leads to the next and ordering matters. One can, for example, not approve a request before receiving it. In process analytics, we exploit this knowledge and we mine data to obtain explicit insights into these causal relations in the form or Petri nets, C-nets, or other process notations. The research in the my research group continues to expand outward from a classical situation of data with clear case notions in the context of explicitly structured processes to a broad, multi-faceted field, where processes are less structured or consist of many interacting artifacts and where case notions in data become more fluid or are complex, multi-dimensional networks.