« Back to home

Causality and Machine Learning

I consider this talk by Bernhard Schoelkopf at the Royal Society, London on causal relationships very interesting. He motivates the use of statistical methods for machine learning and then dives into the fascinating topic of causal modeling.

One of the most exciting concepts during this talk is causal reasoning, which is used to explain causation. One attempt to approach causation is based on the spatio-temporal
characteristics of events
, essentially those triggered by other mechanisms. On the other hand, another school explains causation as the connection between correlated events [1] .

In recent years, an due to the groundbreaking works of Judea Pearl, the science of modeling causality has taken new directions. The concepts to interventions and counterfactuals have been introduced to represent causal models. I like this example mentioned in [2]:

Both smoking and having yellow nicotine-stained teeth are associated with lung cancer. So if you see yellow teeth, you can predict the presence of cancer. However, only a causal account of the disease leads to the correct prediction that a tooth-brushing intervention will have no effect on the cancer rate,
while a smoking-prevention intervention will. Similarly, causal knowledge supports counterfactual claims. A causal account of cancer will also tell you that, had smoking been discouraged in the past, many lives would have been saved

As explained here, the speaker introduces two ideas on how to infer a causal model, these are:

  • The mechanism and the cause are to be kept independent.
  • Applying hard restrictions to the functional model to simplify the scope of its relations.

The applications for this kind of research are countless, besides its practical purposes it is a fascinating topic that unfortunately is not very well understood. Hopefully, more talks like this will become available to wider audiences.


[1] Learning to Learn Causal Models, Kemp et al., 2010

[2] Counterfactuals 2nd Edition, David K. Lewis, 1973


comments powered by Disqus