Non-local, semantic compiler transformations such as Automatic Differentiation and certain Bayesian Inference algorithms pose particular challenges for dynamic dispatch systems, where non-local information may not necessarily be computable at compile time. In this talk, I will describe how we approach these problems in Julia while retaining performance scalability all the way from completely dynamic information situations to the semi-static case. Along the way, I will introduce optical constructions, a recent result from Category Theory that provides a useful guide when choosing abstractions for this class of transformations.

The discussion following this talk will be moderated by Simon Peyton Jones.

Keno Fischer is a core developer of the Julia programming language and co-founder and CTO at Julia Computing. Keno has been working on Julia for most of the past 8 years, contributing major parts of the compiler, REPL and binary package management infrastructure, as well as the Julia Machine Learning stack. In his research, Keno likes to push the boundaries of possibilities for programming systems, from exascale computation to practical systems for homomorphic encryption. Keno holds an AM degree in Physics from Harvard University and was recognized by Forbes as one of their “30 under 30”.