Bayesian Methods for Causal Discovery

The problem I am interested in is using machine learning to simulate the scientific process of
seeing data, forming hypotheses, and choosing experiments to evaluate the hypotheses. As
experiments involve interacting with variables in a system (interventions), which changes the
system from when it is just being observed, this requires using machine learning to model
changes in distribution. A method that can do the above is necessary to understand complex
systems with a large number of variables, unknown interactions, and for which lots of data is
available. For example, recent advances in technology have made lots of observational data
available about gene expressions, as well as the ability to carry out experiments.
Analysing this data to learn the effects each variable has on others is prohibitive due to the
large number of variables and noisy data, however understanding this can have massive impact
on areas such as drug design and personalised medicine.

General area Machine Learning