PI Institution: Clopinet
The objective of this research is to devise a virtual laboratory environment to enable researchers in computational intelligence to propose and study new methods of causal discovery, the problem of attributing causes to effects. Causal discovery is essential to scientific discovery and decision making in domains as diverse as medicine, epidemiology, ecology, and economy. The need for assisting policy making and the availability of massive amounts of data puts new requirements on the design of effective methods to automate causal discovery. The approach is to first organize a challenge around a number or real and semi-artificial tasks to refine evaluation metrics, then design and implement an interactive web-based environment facilitating virtual experiments on artificial systems.
Intellectual merit: Standard benchmarks are needed to foster scientific progress, but the design of a good causal discovery benchmark platform, not biased in favor a particular model or approach, is not trivial. Two key elements distinguish this proposal: (1) the interactive aspect of the platform allowing researchers to emulate experimentation by placing queries on the systems; (2) the dynamic aspect of the platform acting as a repository of new problems and proposed solutions.
Broader impact: Challenges with open participation over the Internet, provide a good opportunity for researchers of any origin to demonstrate excellence and quickly make themselves known. Advancing the methodology for reliably determining causal relationships would have a great socio-economical impact, e.g. by improving policy making. The subject of causal discovery is particularly pedagogical to develop the logic of scientific discovery in students.
