On page 98 of Jet Substructure at the Large Hadron Collider: A Review of Recent Advances in Theory and Machine Learning the author writes;

Redacted phase space: Studying the distribution of inputs and the network performance after conditioning on standard physically-inspired features can help to visualize what new information the network is using from the jet. Training the network on inputs that have been conditioned on specific values of known features can also be useful for this purpose.

I cannot find other references to conditioning features. What does that mean?


1 Answer 1


This is conditioning in the sense of conditional probability. The idea is that the authors have some "standard physically-inspired features". They are splitting the data up into bins based on the values of these features, and then training a model for each bin. They are then examining the differences between the models. Usually this is done to learn something about the benefits of using the different features, and about the relationships between features and outputs.

  • 1
    $\begingroup$ That makes sense, do you happen to have an example of that technique? I guess it could give you good ways to make cuts on the input, but I'm not sure what else it would demonstrate. $\endgroup$
    – Clumsy cat
    Nov 14, 2019 at 13:31
  • $\begingroup$ @Clumsycat A lot of good examples are illustrated by Simpson's paradox (en.wikipedia.org/wiki/Simpson%27s_paradox). More generally, this technique is key in establishing causal linkages. $\endgroup$ Nov 14, 2019 at 21:17

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .