Course of causality, taught by Jonas Peters, University of Copenhagen

Recently, most machine learning has focused on pure predictive performance , This is the driving force behind its actual success . The question of causality ( Understand why predictions work ) Left behind to some extent . This model is very important , Because it can help understand which genes cause which diseases , Which policies affect which economic indicators .

 

In the field of causality , We want to know that the system is interfering ( Such as gene knockout experiment ) Reaction under the pressure . These problems go beyond statistical dependence , Therefore, standard regression or classification techniques cannot be used to answer . In this tutorial , You will learn about the interesting problems of causal reasoning and the latest developments in this field . There is no need to know the causal relationship in advance .

 

The first part : We introduce structural causal models and formal intervening distributions . We define causal effects , And show how to calculate them , If the causal structure is known .

 

The second part : We propose three ideas that can be used to infer causal structures from data :(1) Found in data ( Conditions ) independence ,(2) Restricted structural equation model ,(3) Use the fact that the causal model remains unchanged in different environments .

 

The third part : We show how the concept of causality can be used in more classical machine learning problems .

 

The fourth part : Application of machine learning