onetwo12
For me Lisp was much more appropiate and easy to program. Forward chaining and unification is not enough for certain class of algorithms, sometime you need backward chaining, probabilistic inference and many other ideas that are not easily translated in prolog. I remember some discussion about the semantic of clauses. For example misil(ready), is this about the misil is already ready, or that you are going to prepare the misil, etc. I find that concurrency and paralelism in prolog are not easy. In the book "on Lisp" by P. Grahan there is a prolog interpreter in Lisp, this can be interesting for somebody wishing to learn prolog.
Edit,Added, erlang was inspired by prolog, so learning prolog will help you learn erlang, but if you are a ruby type elixir is your best option.
Fermi estimates require you to look for relations in the real world, construct a model of the situation and then iterate improving the model by adding more information. You don't use that process just to give an example of using a rule or solving an equation, instead you emphasize how the problem could be tackle mathematically. So there are more room for open questions, incentives for exploring and suggesting new approaches, in a more relaxed atmosphere creativity can flourish, perhaps maths is more than a single equation written in an old book.
In the retrofitting paper cited in the comments there is a process of smoothing, that is feeding back the message or information to update the states of the graph (example in the modern ai book). It doesn't seem anything new.
Also, chapter three about SVD, is in https://jeremykun.com/2016/04/18/singular-value-decompositio...
and https://jeremykun.com/2016/05/16/singular-value-decompositio...
the advantage is that you have the python code available.
https://jeremykun.com/2015/04/06/markov-chain-monte-carlo-wi...
The book seems to be interesting.