> the model should be consistent with knowledge about the problem [...] which I don't consider the norm in ML at all. I don't think that is so niche. Murphy's

> the model should be consistent with knowledge about the problem [...] which I don't consider the norm in ML at all. I don't think that is so niche. Murphy's vol II, a mainstream book, starts with this quote: "Intelligence is not just about pattern recognition and function approximation. It’s about modeling the world." — Josh Tenenbaum, NeurIPS 2021. Goodman & Tenenbaum have written e.g. https://probmods.org https://probmods.org, which is very much about modeling data-generating processes. The same can be said about large parts of Murphy's book, Lee & Wagenmakers or Lunn et al. (the BUGS book).