Textbook
Causal Discovery: Foundations, Algorithms and Applications (Computer Science Foundations and Applied Logic)Author(s): Luis Enrique Sucar
2025


Causal reasoning is a constant element in our lives as it is in human nature to
constantly ask why. Acting in the world is conceived by human beings as inter-
vening the world, and humans are able to learn and use causal relations. However,
the development of causality as a scientific discipline is relatively recent. Causal
models and causal reasoning had undergone an important development since the
90s of the previous century, which has been defined as a
causal revolution
. Why?
Because causal models have several advantage with respect to associative tech-
niques (such as statistical models and most of those used in machine learning).
First, a causal model provides a deeper understanding of a domain, by identifying
the direct and indirect causes of certain events, which is not necessarily true for
associative models. Secondly, with causal models we can perform other types of
reasoning that are not possible, at least in a direct way, with associative models.
These other inference situations are: (i)
interventions
, in which we want to find the
effects of setting a certain variable to a specific value by an external agent (note
that this is different from observing a variable); and (ii)
counterfactuals
, where we
want to reason about what would have happened if certain information had been
different from what actually happened. Causality is also important for explana-
tory purposes, since an effect can be explained by its causes, regardless of the
correlations it may have with other v ariables.
However, a challenge is how to learn causal models. Traditionally, to build a
causal model we need to perform interventions in the
real
world, such as those in
randomized clinical trails. Performing this type of interventions is costly, and in
some cases unethical or impossible. But we have a lot of
observational
data, that
is data obtained from natural processes, without an explicit intervention. This is a
challenge, as only from observational data we cannot recover, in general, a unique
causal model. There has been an increasing interest of the research community in
learning causal models from observational data, what is known as
causal discovery
,
with important developments in the last few years.
The objective of this book is to present an overview of the main algorithms and
techniques for causal discovery, complementing other books on causality which



mainly focus on causal reasoning. As this is a developing area, we describe in
detail the basic classes of algorithms, and provide references to other variations
and more recent developments. Although the focus is causal discovery, it provides
the necessary background on causal modeling, with emphasis on causal graphical
models; as well as causal reasoning, including interventions and counterfactuals.
Some of the key features of the book are:
•It includes the necessary background material: a review of probability and graph
theory, Bayesian networks, causal graphical models and causal reasoning.
•The main types of causal discovery are covered: learning from observational
data, learning from interventional data, and learning from temporal data.
•It illustrates the application of causal discovery in practical problems, an
important feature for students and practitioners.
•It includes some of the latest developments in the field, such as continuos opti-
mization, causal event networks, causal discovery under subsampling, subject
specific causal models, causal reinforcement learning, etc.
•Each chapter has a set of exercises, including suggestions for research and
programming projects.
As causal discovery is a relatively new field, its applications are just developing.
To illustrate the potential of causal modeling and discovery in many fields, and
to complement the theory with practical applications; the book includes many
examples of the application of causal discovery in a wide range of domains, such
as:
•Eliciting brain effective connectivity.
•Causal discovery of the risk factors for COVID.
•Learning causal models of gene regulatory networks.
•Understanding the factors that affect decisions for the energy transition.
•Predicting the stock market.
•Causal inference in legal texts.
•Causal discovery from images.
Drone navigation incorporating causal models.


Contents