Recently I have been reading papers by Uwe Naumann:
https://www.stce.rwth-aachen.de/people/uwe-naumann
He's one of the senior researchers within the field of AD (Automatic
Differentiation), which I have also been interested in for the past
few years. I quite enjoy reading his papers, although they come from a
very different tradition (numerical computing) than me, and so some
concepts or phrasing are alien to me. It is not that Uwe Naumanns's
writing is funny or poetic or elegant in the way that e.g. is the case
for SPJ's papers - I think he just writes about that nice boundary
between theory (or really, mathematics) and practical matters that I
also enjoy. A lot of the papers read like a person explaining this
neat tool he made, which is a style I quite enjoy.
Another cultural curiosity I found is that Uwe will sometimes talk
about applying AD "by hand". In fact, Uwe also uses a different
definition of AD: *Algorithmic* Differentiation, rather than
*Automatic* Differentiation. The principle is the same
(differentiating arbitrary computer programs via repeated application
of the chain rule), but I think Uwe's perspective is that you could
apply the transformation by hand if you wanted to, although most of
his papers emphasize that this does not scale at all.
Some of his papers that I have read so far are:
* Adjoint Code Design Patterns
* A Matrix-Free Exact Newton Method
* Seasons's Greetings by AD
* Differentiable Scripting
* Adjoint algorithmic differentiation tool support for typical
numerical patterns in computational finance
I also need to get my hands on his book *The Art of Differentiating
Computer Programs*.
